on nonequilibrium statistical mechanics
TRANSCRIPT
Western University Western University
Scholarship@Western Scholarship@Western
Electronic Thesis and Dissertation Repository
8-17-2015 12:00 AM
On Nonequilibrium Statistical Mechanics On Nonequilibrium Statistical Mechanics
Joshua M. Luczak The University of Western Ontario
Supervisor
Wayne C. Myrvold
The University of Western Ontario
Graduate Program in Philosophy
A thesis submitted in partial fulfillment of the requirements for the degree in Doctor of
Philosophy
© Joshua M. Luczak 2015
Follow this and additional works at: https://ir.lib.uwo.ca/etd
Part of the Philosophy of Science Commons
Recommended Citation Recommended Citation Luczak, Joshua M., "On Nonequilibrium Statistical Mechanics" (2015). Electronic Thesis and Dissertation Repository. 3173. https://ir.lib.uwo.ca/etd/3173
This Dissertation/Thesis is brought to you for free and open access by Scholarship@Western. It has been accepted for inclusion in Electronic Thesis and Dissertation Repository by an authorized administrator of Scholarship@Western. For more information, please contact [email protected].
ON NONEQUILIBRIUM STATISTICAL MECHANICS
(Thesis Format: Monograph)
by
Joshua Luczak
Graduate Program in Philosophy
A thesis submitted in partial fulfilment
of the requirements for the degree of
Doctor of Philosophy
The School of Graduate and Postdoctoral Studies
The University of Western Ontario
London, Ontario, Canada
c© Joshua Luczak, 2015
Abstract
This thesis makes the issue of reconciling the existence of thermodynami-
cally irreversible processes with underlying reversible dynamics clear, so as to
help explain what philosophers mean when they say that an aim of nonequi-
librium statistical mechanics is to underpin aspects of thermodynamics.
Many of the leading attempts to reconcile the existence of thermodynam-
ically irreversible processes with underlying reversible dynamics proceed by
way of discussions that attempt to underpin the following qualitative facts:
(i) that isolated macroscopic systems that begin away from equilibrium spon-
taneously approach equilibrium, and (ii) that they remain in equilibrium for
incredibly long periods of time. These attempts standardly appeal to phase
space considerations and notions of typicality. This thesis considers and eval-
uates leading typicality accounts, and, in particular, highlights their limita-
tions. Importantly, these accounts do not underpin a large and important set
of facts. They do not, for example, underpin facts about the rates in which
systems approach equilibrium, or facts about the kinds of states they pass
through on their way to equilibrium, or facts about fluctuation phenomena.
To remedy these and other shortfalls, this thesis promotes an alternative, and
arguably more important, line of research: understanding and accounting for
the success of the techniques and equations physicists use to model the be-
haviour of systems that begin away from equilibrium. Accounting for their
success would help underpin not just the qualitative facts the literature has
focused on, but also many of the important quantitative facts that typicality
accounts cannot.
ii
This thesis also takes steps in this promising direction. It outlines and
examines a technique commonly used to model the behaviour of an interesting
and important kind of system: a Brownian particle that’s been introduced to
an isolated homogeneous fluid at equilibrium. As this thesis highlights, the
technique returns a wealth of quantitative and qualitative information. This
thesis also attempts to account for the success of the model and technique,
by identifying and grounding the technique’s key assumptions.
Keywords: statistical mechanics, thermodynamics, foundations of thermo-
dynamics, reversible, time-reversal non-invariance, laws of thermodynamics,
typicality, approach to equilibrium, Brownian particle, Langevin dynamics.
iii
Acknowledgements
Very special thanks to my supervisor Professor Wayne Myrvold whose
knowledge, helpful comments and advice, encouragement, and support, has
helped, and pushed me, at each stage of my philosophical career, to produce
the very best work I am capable of. He has been a model philosopher and a
true inspiration.
Thanks to Professors Robert DiSalle, Carl Hoefer, Markus Muller, Stathis
Psillos, Chris Smeenk, and David Wallace for their helpful comments, advice,
encouragement, and support. Thanks to Professor William Harper for his
enthusiasm. His love of philosophy and science is infectious. And thank you
to all of my other teachers, and to my fellow graduate students at Western.
I have learnt a great deal from you.
Thanks to audiences at Ghent, Helsinki, Notre Dame, Oxford, the Uni-
versity of California Irvine, and Western, for their comments on work related
to this project.
Thanks to my family, whose support is greatly appreciated. In particular,
thank you to my great aunt, Erica, whose interest in my studies has always
encouraged me to reach higher. Special thanks also goes to my parents, Fiona
and Maciek, who offer unwavering love and support.
Finally, a very big thanks goes to my partner, Nanette, whose love, sup-
port, enthusiasm, advice, encouragement, and sense of adventure has made
my greatest achievements possible. No victories are as sweet as the ones I
share with you.
iv
Contents
Abstract ii
Acknowledgements iv
List of Figures vii
List of Tables vii
1 Introduction 1
1.1 Chapter Summaries . . . . . . . . . . . . . . . . . . . . . . . . 3
1.1.1 Chapter 2 . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.1.2 Chapter 3 . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.1.3 Chapter 4 . . . . . . . . . . . . . . . . . . . . . . . . . 7
2 On Some of the Aims of Nonequilibrium
Statistical Mechanics 9
2.1 Classical Thermodynamics . . . . . . . . . . . . . . . . . . . . 14
2.1.1 Time-Reversal Invariance and “Reversibility” . . . . . 15
2.1.2 The Minus First Law . . . . . . . . . . . . . . . . . . . 20
2.1.3 The Zeroth Law . . . . . . . . . . . . . . . . . . . . . . 23
2.1.4 The First Law . . . . . . . . . . . . . . . . . . . . . . . 24
2.1.5 The Second Law . . . . . . . . . . . . . . . . . . . . . 26
2.2 Summary of Results . . . . . . . . . . . . . . . . . . . . . . . 38
2.3 Not Strict Interpretations, Suitable Analogues . . . . . . . . . 41
3 An Evaluation of Foundational Accounts 44
3.1 Boltzmann’s Early Work . . . . . . . . . . . . . . . . . . . . . 47
v
3.1.1 Boltzmann’s H-theorem . . . . . . . . . . . . . . . . . 49
3.1.2 The Kac Ring . . . . . . . . . . . . . . . . . . . . . . . 51
3.2 Typicality Accounts . . . . . . . . . . . . . . . . . . . . . . . . 57
3.2.1 Some Background . . . . . . . . . . . . . . . . . . . . . 59
3.2.2 The Dominance View . . . . . . . . . . . . . . . . . . . 61
3.2.3 The Unspecified Dynamical View . . . . . . . . . . . . 65
3.2.4 The Ergodic View . . . . . . . . . . . . . . . . . . . . . 68
3.2.5 The Epsilon Ergodic View . . . . . . . . . . . . . . . . 72
3.3 Some General Concerns About Typicality Accounts . . . . . . 74
3.3.1 Typicality Measures . . . . . . . . . . . . . . . . . . . 75
3.3.2 Extending Typicality Results . . . . . . . . . . . . . . 80
3.4 The Limitations of Typicality Accounts . . . . . . . . . . . . . 82
3.5 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
4 Another Way to Approach the Approach to Equilibrium 88
4.1 Modelling a Brownian Particle’s Behaviour . . . . . . . . . . . 90
4.2 Some Comments on the Approach . . . . . . . . . . . . . . . . 100
4.2.1 Constructing a Langevin Equation . . . . . . . . . . . 101
4.2.2 The Time-Reversal Non-Invariance of the
Langevin Equation . . . . . . . . . . . . . . . . . . . . 104
4.2.3 Why Does the Langevin Equation Work? . . . . . . . . 107
4.3 Motivating the Collision Assumption . . . . . . . . . . . . . . 109
4.4 A Summary and a Suggestion . . . . . . . . . . . . . . . . . . 112
5 Conclusion 115
Bibliography 118
Curriculum Vitae 130
vi
List of Figures
2.1 A heat engine . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
2.2 A Carnot engine . . . . . . . . . . . . . . . . . . . . . . . . . 29
2.3 A Carnot engine operating in reverse . . . . . . . . . . . . . . 30
2.4 A heat engine running a Carnot engine in reverse . . . . . . . 30
2.5 The four stages of a Carnot Cycle . . . . . . . . . . . . . . . . 33
2.6 A violation of the Kelvin statement . . . . . . . . . . . . . . . 37
2.7 A violation of the Clausius statement . . . . . . . . . . . . . . 37
3.1 A Kac ring . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
3.2 The dominance of the equilibrium macrostate . . . . . . . . . 63
4.1 The mean squared velocity of a Brownian particle as a function
of time . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
List of Tables
4.1 Observed displacements of a spherical particle immersed in
water . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
vii
1
Chapter 1
Introduction
Contemporary philosophical discussions of statistical mechanics primarily
focus on two issues: securing a foundation for thermodynamics, and inter-
preting statistical mechanical probabilities. This thesis speaks to the first of
these issues.1 In particular, it speaks to aspects of the foundational debate
that make contact with nonequilibrium statistical mechanics.
This thesis has several aims. One of the aims is to help explain what
philosophers mean when they say that an aim of nonequilibrium statistical
mechanics is to account for aspects of thermodynamics. While there are a
number of aspects of thermodynamics whose justification relies on nonequi-
librium statistical mechanics, philosophers have tended to focus on the is-
sue of reconciling the existence of thermodynamically irreversible processes
with underlying reversible dynamics. Interestingly, and perhaps somewhat
surprisingly, this goal is expressed in different ways throughout the litera-
1For good introductions and discussions on how we should understand the probabil-ity distributions introduced into statistical mechanics, see Sklar (1995), Frigg (2008b),Myrvold (2012), and Myrvold (Forthcoming).
2
ture. Or rather, what’s seen as central to achieving this goal differs across
authors—whether this is intended or not. To an outsider, these differences
may suggest that there is a disagreement or confusion in the literature. It is
an aim of this thesis to make the issue of reconciling the existence of thermo-
dynamically irreversible processes with underlying reversible dynamics more
clear, and to explain why there is neither a disagreement or confusion in the
literature. This is done so as to satisfy the more general aim of helping to ex-
plain what philosophers mean when they say that an aim of nonequilibrium
statistical mechanics is to underpin aspects of thermodynamics.
Many of the leading attempts to reconcile the existence of thermodynam-
ically irreversible processes with underlying reversible dynamics proceed by
way of discussions that attempt to underpin the following qualitative facts:
(i) that isolated macroscopic systems that begin away from equilibrium spon-
taneously approach equilibrium, and (ii) that they remain in equilibrium for
incredibly long periods of time. These attempts standardly appeal to phase
space considerations and notions of typicality. It is an aim of this thesis to
consider and evaluate leading typicality accounts, and, in particular, to high-
light their limitations. Importantly, these accounts do not underpin a large
and important set of facts. They do not, for example, underpin facts about
the rates in which systems approach equilibrium, or facts about the kinds of
states they pass through on their way to equilibrium, or facts about fluctua-
tion phenomena. To remedy these and other shortfalls, this thesis promotes
an alternative, and arguably more important, line of research: understanding
and accounting for the success of the techniques and equations physicists use
to model the behaviour of systems that begin away from equilibrium. Ac-
3
counting for their success would help underpin not just the qualitative facts
the literature has focused on, but also many of the important quantitative
facts that typicality accounts cannot.
This thesis also takes steps in this promising direction. It is an aim of
the fourth chapter to outline, examine, and ground the success of a tech-
nique commonly used to model the approach to equilibrium and subsequent
behaviour of a Brownian particle that’s been introduced to an isolated homo-
geneous fluid at equilibrium. Interestingly, the model is easily generalised,
and so is applicable to a wide variety of systems. This chapter attempts
to account for the success of the model, by identifying and grounding the
technique’s key assumptions.
It’s standard for philosophical discussions of statistical mechanics to take
place within a classical mechanical framework. This thesis follows this con-
vention and eschews quantum mechanical considerations. While the following
view will not be defended in this thesis, it is the view of the author that clas-
sical statistical mechanics needs no help from quantum mechanics to make
sense; classical statistical mechanics can stand on its own.
1.1 Chapter Summaries
1.1.1 Chapter 2
A number of authors claim that an aim of statistical mechanics is to provide
a suitable foundation for thermodynamics. With respect to the role nonequi-
librium statistical mechanics plays in the foundational project, philosophers
have tended to focus on reconciling the existence of thermodynamically irre-
4
versible processes with underlying reversible dynamics. Interestingly, what’s
seen as central to this task differs across authors. Or rather, and more ac-
curately, many authors use distinct concepts when they outline what they
see as central to reconciling the tension between these macroscopic processes
and the underlying dynamics of those systems that give rise to them. Impor-
tantly, these differences may indicate, to those unfamiliar with the literature,
that there is either a confusion or disagreement in the literature. This chapter
highlights these differences. This is done via a discussion of various reversibil-
ity concepts and several laws of thermodynamics. It also explains why these
differences, in fact, do not amount to there being a confusion or disagree-
ment in the literature, and why well informed authors needn’t (and probably
don’t) worry about eliminating these differences. This chapter encourages
thinking of these different statements as short, approximate statements that
refer to a collection of relevant ideas central to the reconciliation issue.
This chapter, by way of making the reconciliation issue more clear, helps
explain what philosophers mean when they say that an aim of statistical
mechanics is to underpin aspects of thermodynamics. And this, in turn,
contributes to this thesis’s aim of encouraging that we pursue an alternative
line of research within the philosophical literature on statistical mechanics:
to understand and account for the success of the techniques and equations
physicists use to model the behaviour of systems that begin away from equi-
librium. By better understanding the goals philosophers have traditionally
set themselves, we’ll not only be able to better situate, appreciate, and evalu-
ate their attempts to achieve them, but we’ll also better appreciate how this
proposed line of research fits with contemporary discussions of statistical
5
mechanics.
1.1.2 Chapter 3
Chapter two discusses the reconciliation of thermodynamically irreversible
processes with underlying reversible dynamics with reference to thermody-
namics. This chapter discusses the issue from the perspective of statistical
mechanics.
One of the aims of this chapter is to outline and review leading attempts
to address the reconciliation issue. These attempts are typically part and
parcel of works that aim at accounting for why macroscopic systems approach
equilibrium and why they remain in equilibrium for incredibly long periods
of time. Almost all of the attempts discussed in this chapter appeal to a
notion of typicality.
This chapter discusses the most pressing problems with typicality ac-
counts and, where possible, discusses their most promising solutions. It is
argued that they have severe limitations, even when we overlook their prob-
lems. It is argued that they do not underpin a large and important set of
facts that concern the behaviour of systems that begin away from equilib-
rium. While they may account for why isolated macroscopic systems ap-
proach equilibrium and why they remain in equilibrium for incredibly long
periods of time, they do not, among other things, underpin facts about the
rates in which systems approach equilibrium, or about the kinds of states
they pass through on their way to equilibrium, or about fluctuation phenom-
ena. They do not help us form expectations about these aspects of a system’s
behaviour or help justify the expectations we may already have about them,
6
having formed them on the basis of experience.
This chapter claims that the limitations of typicality accounts are a symp-
tom of what they are aiming at. By focusing on recovering aspects of thermo-
dynamics, those contributing to the foundational project have merely been in
the business of articulating microphysical accounts of irreversible macroscopic
behaviour that recover a few qualitative facts. To remedy these shortfalls,
this chapter promotes pursuing a different line of research; one in which we
attempt to understand and ground the success of the techniques and equa-
tions physicists use to model the behaviour of systems that begin away from
equilibrium. It is claimed that accounting for their success would help un-
derpin not just the qualitative facts the literature has focused on, but also
many of the important quantitative facts that typicality accounts cannot.
This chapter is composed of five sections. The first section outlines and
discusses the essence of what many people think of as the first serious attempt
to account for the behaviour of systems that begin away from equilibrium.
This approach is based on Ludwig Boltzmann’s early work on statistical me-
chanics, when he first offers what has come to be known as Boltzmann’s
equation and H-theorem. As this section highlights, this work gives rise to
one of the very issues that is at the centre of contemporary discussions of
statistical mechanics: the reconciliation of thermodynamically irreversible
processes with underlying reversible dynamics. The second section outlines
and discusses leading ways of accounting for the behaviour of systems that
begin away from equilibrium that also addresses the tension between their
macrolevel and microlevel descriptions. These approaches are inspired by,
and build on, Boltzmann’s attempts (circa 1877) to resolve this tension and,
7
once again, account for the behaviour of systems that begin away from equi-
librium. As we’ll see, each one incorporates a notion of typicality. The
second section also considers worries specific to each kind of typicality ac-
count discussed. More general concerns are discussed in the third section.
Both sections also discuss, where possible, the most promising solutions to
these concerns. The fourth section highlights and discusses the limitations
of typicality accounts. It also encourages those working in the field to be-
gin work on understanding and grounding the success of the techniques and
equations physicists use to model the behaviour of systems that begin away
from equilibrium. The fifth section summarises the material presented in this
chapter.
1.1.3 Chapter 4
Chapter three ends by highlighting the limitations of typicality accounts and
by suggesting that we pursue a line of research aimed at properly under-
standing and accounting for the success of the techniques physicists use to
model the behaviour of systems that begin away from equilibrium.
This chapter takes steps in this promising direction. It outlines and ex-
amines a technique commonly used to model the behaviour of an interesting
and important kind of system. More accurately, this chapter outlines and
examines a technique commonly used to model the behaviour of a Brownian
particle that’s been introduced to an isolated homogeneous fluid at equi-
librium. As we’ll see, the technique returns a wealth of quantitative and
qualitative information. This chapter also attempts to account for the suc-
cess of the model and technique, by identifying and grounding the technique’s
8
key assumptions.
This chapter is composed of four sections. The first section details a
way of modelling the Brownian particle’s approach to equilibrium and sub-
sequent behaviour. The approach takes its cue from the theory of Brownian
motion. It appeals to Langevin dynamics. As this section highlights, the
described technique generates a collection of interesting equations that track
important aspects of the system’s behaviour. These equations provide us
with the resources to answer the kinds of questions commonly asked about
the behaviour of this kind of system. The second section expands and com-
ments on several aspects of the approach. It also identifies the technique’s
key assumptions and discusses their motivation. This section also begins a
discussion about what would have to be true, or be at least approximately
true, at the microscopic level, in order to justify the use of these assump-
tions, and to account for the success of the equations they help generate.
The third section attempts to motivate a crucial microphysical fact, which,
it is claimed, justifies the use of the technique’s key assumptions, grounds
the success of the technique, and underpins the facts the equations it helps
generate track. The fourth section summarises the details of this chapter. It
also contains some suggestions about the direction of future investigations.
9
Chapter 2
On Some of the Aims of
Nonequilibrium Statistical
Mechanics
A number of authors claim that the aim of statistical mechanics is to provide
a suitable foundation for thermodynamics.1,2 For example, in a recent review
article Roman Frigg writes:
Thermodynamics (TD) correctly describes a large class of phe-
nomena we observe in macroscopic systems. The aim of statistical
mechanics is to account for this behaviour in terms of the dynam-
ical laws governing the microscopic constituents of macroscopic
systems and probabilistic assumptions. (Frigg 2008b: p.99)
1See, for example, Sklar (1995: p.3), Lebowitz (1999: p.346), Uffink (2007: p.923), andFrigg and Werndl (2012b: pp.99-100).
2This observation has also recently been made by Wallace (2013).
10
Others, perhaps more cautiously, claim that it is a central aim of sta-
tistical mechanics to provide a foundation for thermodynamics. Katinka
Ridderbos (2002: p.66), for example, says that
One of the cardinal aims of the theory of statistical mechanics is
to underpin thermodynamic regularities by a theory formulated
in terms of the dynamical laws governing the motion of the mi-
croscopic constituents of a thermodynamic system.
While Craig Callender (2001: p.540) writes,
Kinetic theory and statistical mechanics are in part attempts to
explain the success of thermodynamics in terms of the basic me-
chanics.
As one might have expected, the consensus among philosophers and philo-
sophically minded physicists (which will hereafter collectively be referred to
as philosophers) is that it is an aim of nonequilibrium statistical mechanics
to underpin certain aspects of thermodynamics. While there are a number
of aspects of thermodynamics whose justification relies on nonequilibrium
statistical mechanics, philosophers have tended to focus on the issue of rec-
onciling the existence of thermodynamically irreversible processes with un-
derlying reversible dynamics. Interestingly, what’s regarded as central to this
task differs across authors. Authors sometimes claim that:
11
G1. It is a goal of nonequilibrium statistical mechanics to account for ther-
modynamic behaviour, where we say that an isolated system is be-
having thermodynamically if it’s in equilibrium or if it’s spontaneously
approaching equilibrium.3
G2. It is a goal of nonequilibrium statistical mechanics to account for irre-
versible macroscopic behaviour.4
G3. It is a goal of nonequilibrium statistical mechanics to show how a time-
reversal non-invariant thermodynamics emerges from a time-reversal
invariant microphysics.5
G4. It is a goal of nonequilibrium statistical mechanics to provide a micro-
physical justification for the Second Law of thermodynamics.6
Importantly, these goals are distinct. They are, however, closely related.
For one thing, G1 and G2 do not make explicit reference, like G3 and G4,
to some microphysics. It’s clear, however, from the context in which they
appear, that what authors mean when they say that it’s a goal of nonequi-
librium statistical mechanics to account for thermodynamic behaviour, or
irreversible macroscopic behaviour, is that it’s a goal to provide a micro-
physical underpinning of these concepts.
3See, for example, Lazarovici and Reichert (2014: p.3).4See, for example, Frigg (2011: p.77).5See, for example, Wallace (2013: p.1). Actually, Wallace (2013: p.1) describes some-
thing weaker than this. More accurately, he describes the goal as being one in whichwriters have tried to show how the apparent time-irreversibility of thermodynamics canbe reconciled with the apparent time-reversibility of microphysics. Since this thesis hasconfined itself to talking about classical mechanics and classical thermodynamics, it seemsreasonable to drop the “apparent” qualifier.
6See, for example, Frigg (2008a: p.670) and Lazarovici and Reichert (2014: p.4).
12
One of the aims of this chapter is to help explain what philosophers mean
when they say that an aim of nonequilibrium statistical mechanics is to ac-
count for certain aspects of thermodynamics. Another is to highlight the
ways in which G1-G4 are distinct, and to highlight the ways in which they
are related. A third is to explain why these differences are negligible with re-
spect to the reduction of thermodynamics to statistical mechanics. A fourth
is to explain why these distinct but related ways of expressing what’s central
to reconciling the tension between macrolevel and microlevel descriptions of
relevant phenomena do not reveal either a confusion or disagreement in the
literature. All of these aims are related. By discussing the ways in which
these goals are distinct, related, but ultimately whose differences are negligi-
ble with respect to the reduction of thermodynamics to statistical mechanics,
this chapter hopes to make the issue of reconciling the existence of thermo-
dynamically irreversible processes with underlying reversible dynamics more
clear. If it’s successful, then this chapter will have helped articulate what it
is those attempting to underpin aspects of thermodynamics using nonequi-
librium statistical mechanics are actually trying to achieve. The attempt to
satisfy each of these aims also contributes, importantly, to this thesis’s larger
aim of encouraging the pursuit of a different line of research. By better un-
derstanding thermodynamics, its relationship to statistical mechanics, and
the facts philosophers have traditionally tried to underpin using statistical
mechanics, we’ll be able to better situate, and appreciate, this thesis’s pro-
motion of understanding and accounting for the success of the techniques
physicists use to model the behaviour of systems that begin away from equi-
librium.
13
While it’s helpful for this thesis’s narrative that this chapter proceeds
by way of a discussion about some distinct but roughly equivalent claims
about what’s central to reconciling the existence of irreversible macroscopic
processes with an underlying reversible dynamics, it also serves the purpose
of later clarifying the relationship between thermodynamics and statistical
mechanics. As a later section explains, the general expectation, among those
working on the reduction of thermodynamics to statistical mechanics, is that
the laws and concepts of thermodynamics appear in statistical mechanics as
complicated, approximate statements that are true under certain conditions.
Importantly, it is for this reason that these distinct but related claims do
not properly indicate that there is either a confusion or disagreement in the
literature.
This chapter consists of three sections. The first section introduces and
discusses some concepts, and some thermodynamic laws, that are relevant
to ways of interpreting G1-4. They are also relevant for understanding the
ways in which these goals are distinct but related. Since much of the founda-
tional literature is concerned with recovering classical thermodynamics, this
discussion focuses on classical thermodynamic concepts and classical ther-
modynamic laws—though often in their modern guise. This is distinguished
from the less orthodox, more formal, axiomatic treatments of thermodynam-
ics that have appeared since the early 1900’s, and which are associated with
Constantin Caratheodory, and Elliott Lieb and Jakob Yngvason.7 The re-
sults of this discussion are summarised in the second section of this chapter.
7See Caratheodory (1909), and Lieb and Yngvason (1999). Other axiomatic approacheshave been offered by Robin Giles (1964), John Boyling (1972), and Josef-Maria Jauch(1972, 1975), among others.
14
The third section explains why the differences between G1-4, and goals sim-
ilar to them, are negligible with respect to the reduction of thermodynamics
to statistical mechanics, and, relatedly, why these differences do not point
to either a confusion or disagreement in the literature. It also explains why
authors contributing to the foundational discussion needn’t worry, and prob-
ably don’t worry, about eliminating these differences.
2.1 Classical Thermodynamics
Classical thermodynamics (which is sometimes called orthodox thermody-
namics) was developed around 1850. It’s usually associated with authors
such as Rudolf Clausius, Lord Kelvin (William Thomson), and Max Planck—
though some trace its origins back to Sadi Carnot’s work on heat engines.8
Thermodynamics characterises macroscopic systems in terms of macroscopi-
cally measurable quantities, e.g. temperature, pressure, volume, etc. It also
describes changes in them in terms of heat and work exchanges with an en-
vironment. The theory rests on a set of fundamental laws. These laws are
intended to be independent of any particular hypothesis concerning the mi-
croscopic constitution of macroscopic systems. They have traditionally been
understood as generalised statements of experimental facts.
This section outlines and discusses four thermodynamic laws. They are
8See Mendoza and Carnot (1960). Most authors consider classical thermodynamicsto have emerged around 1850, in works that attempted to recast Carnot’s theorem (seesection 2.1.5)—which was originally expressed using terms familiar to the caloric theoryof heat—in what we now call classical thermodynamic terms. Other early, important,but lesser known contributors to classical thermodynamics are Emile Clapeyron, WilliamRankine, and Ferdnand Reech. See Uffink (2001, 2007) for more on the interesting historyof thermodynamics.
15
known as the Minus First Law, the Zeroth Law, the First Law, and the Second
Law of thermodynamics. Interestingly, only two of these laws—the First and
Second—were stated as laws in original presentations of the theory. While
many modern thermodynamics textbooks include the Zeroth Law, explicit
presentations of the Minus First Law as a law are, at present, only found
in philosophical works. A number of modern textbooks do however include
a Third Law of thermodynamics. Unlike the other laws, the Third Law
is not relevant to the aims of this chapter, so it will not form part of the
discussion.9 The discussion will focus on highlighting aspects of the theory
that bear on G1-4. This will be achieved mostly through a discussion of
thermodynamic laws. This section will also highlight, along the way, some
of the relations between G1-4. But before turning to a discussion of these
laws, it is important that we first disentangle and discuss some important
and closely related concepts.
2.1.1 Time-Reversal Invariance and “Reversibility”
The state of a system is typically represented by a point in some state space
Ω. For example, the thermodynamic state of a system is typically represented
by a point in a state space characterised by a small number of macroscop-
ically measurable parameters.10 The same system can also be represented
in classical phase space, a 6-n-dimensional space in which each point repre-
sents the position and momentum (xi,pi) of every particle constituting the
9See Kardar (2007: Sec.1.10) for an introduction and discussion of the Third Law.10The thermodynamic state of a system need not be characterised by directly macro-
scopically measurable parameters. It can, for example, be characterised using energy andentropy, which have to be measured indirectly via the First and Second laws.
16
system. Unlike thermodynamic state spaces, points in phase space represent
unique microstates. As a consequence, they provide a microphysical, and
hence more detailed, picture of the system’s state.
A state history is a trajectory through state space. That is, a mapping
σ : I ⊆ R→ Ω, for some time interval I.
Systems are governed by dynamical laws. They enable us to distinguish
a set D of dynamically possible trajectories from kinematically possible tra-
jectories.
For any time t0, we can define a reflection of the time axis around t0 by
t→ tT = t0 − (t− (t0)). (2.1)
Standardly, we take t0 = 0, so that tT = −t.
We can talk about a state’s time-reversal. For a single classical particle,
the time-reversal operation
(x,p)→ (x,−p) (2.2)
returns time-reversed states. For systems that are constituted by many
classical particles, the relevant time-reversal operation reverses each particle’s
momentum.
We can also talk about the time-reversal invariance of physical laws and
theories. Given a time reflection (i.e. (2.1)) and a state reversal operation
(such as (2.2)), whose general form we write as
ω → ωT , (2.3)
17
we can define an operation that reverses state histories. Define the
history-reversal operation
σ → σT (2.4)
by
σT (t) = σ(tT )T . (2.5)
So then, if σ includes a sequence of states . . . σ(t1), σ(t2), σ(t3), σ(t4),
. . ., then the time-reversed history includes a sequence of states . . . σ(tT4 )T ,
σ(tT3 )T , σ(tT2 )T , σ(tT1 )T , . . . .
A physical theory is said to be time-reversal invariant (or symmetric un-
der time-reversal) if and only if, whenever a state history σ is dynamically
possible, the time-reversed state history σT is also dynamically possible. An-
other way of putting this is to say that the theory is time-reversal invariant
if and only if DT ⊆ D. A theory is said to be time-reversal non-invariant (or
asymmetric under time-reversal) if and only if it’s not time-reversal invari-
ant. I.e. if there exists at least one state history that is dynamically possible
whose time-reversed history is not.11
Notice that it’s the form of a theory’s dynamical laws, given a state rever-
sal operation, that determines whether or not the theory is time-reversal in-
variant. This is why authors sometimes speak of the time-reversal invariance
of physical laws rather than the time-reversal invariance of physical theories,
11Asymmetry should not be confused with the stronger notion of antisymmetry. A phys-ical theory is said to be antisymmetric under time-reversal if and only if every dynamicallypossible state history has a time-reversed history that is not dynamically possible.
18
and why they sometimes use the expression “time-reversal invariant theory”
interchangeably with “time-reversal invariant dynamical laws”.
“Reversible” is a term that has several meanings. Sometimes it’s used to
refer to time-reversal invariant processes. That is, dynamically possible state
histories whose time-reverse state histories are also dynamically possible.
This use most commonly appears in philosophy of science and philosophy
of physics literature. In these contexts, “irreversible” is used to refer to
time-reversal non-invariant processes. That is, dynamically possible state
histories whose time-reverse state histories are not, according to the theory,
dynamically possible.
In other situations, “reversible” is used to mean, what Jos Uffink
(2001: p.316) has called, recoverable. Experience suggests that in many cases
the transition, by some process, from an initial thermodynamic state si to a
final thermodynamic state sf , cannot be fully undone. The expression “fully
undone” is meant to indicate not just a return of the system to its initial
thermodynamic state but also any auxiliary system with which it interacted,
without any additional change associated therewith. Opening a bottle of
perfume in a well ventilated room is an example of an irreversible processes
in this sense. In this and other such cases, there is no process which starts
off from the final state sf that restores the initial state si, completely. As
Uffink (2001: pp.316-317) notes, this concept of reversibility differs from the
previous one in at least three respects. First, the only thing that matters
for reversibility understood as recovery is a return of the initial thermody-
namic state. We needn’t specify a history-reversal operation that ensures
the composite system pass through its reversed sequence of states. In this
19
respect, this notion of reversibility is weaker than reversibility understood as
time-reversal invariance. And irreversibility understood as irrecoverability
is a logically stronger notion than irreversibility understood as time-reversal
non-invariance. A second difference is its emphasis on complete recovery. A
complete recovery involves not only the return of the system to its initial
state but also the return of any auxiliary system that it interacted with. Re-
versibility understood as time-reversal invariance is simply concerned with
states of the system, and not with the states of systems it interacted with as
well. A third difference concerns the concept of possibility that’s implicitly
invoked. Here, the intended notion of possibility is usually tied to the means
available, to beings like us, with our epistemic and physical limitations, to
recover the initial thermodynamic state. Irreversible processes, then, are pro-
cesses that result in some thermodynamic state from which it is not possible,
given certain epistemic and physical limitations, to recover the initial ther-
modynamic state. This differs from reversibility understood as time-reversal
invariance in that in that case one is simply making a claim about what is
dynamically possible, irrespective of further limitations.
More commonly, however, the word “reversible” is used to mean what
more careful authors mean when they say that certain processes are quasi-
static and reversible. Thermodynamics can be thought to distinguish between
two kinds of processes. First, there are ones that take place gently, with
no turbulence or friction. Then there are all other processes. Quasi-static
processes are processes of the first kind. They are processes that are carried
out so slowly that the system is, at every moment, effectively in equilibrium.
Reversible processes in this context—that is, when they are distinguished
20
from quasi-static processes—are understood as they were in the previous
sense: as processes that occur to systems that result in thermodynamic states
from which it’s possible to recover their initial thermodynamic state along
with the initial thermodynamic state of any auxiliary system with which it
has interacted with, and without there being any other changes associated
therewith.
We can already see from the reversibility concepts outlined above that
there are a number of ways to interpret G2. Importantly, it should be
clear that there are ways of interpreting G2 that are distinct from ways of
interpreting G3.
Having highlighted some things that could be meant by G2 and G3, we
turn now to the task of identifying where these reversibility concepts make
contact with the laws of thermodynamics. We’ll also look to see where goals
such as G1 and G4 make contact with the theory, and with reversibility
concepts.
2.1.2 The Minus First Law
The Minus First Law first appeared in name in Harvey Brown and Jos
Uffink’s 2001 article “The Origins of Time-Asymmetry in Thermodynam-
ics: The Minus First Law”. Earlier authors, however, both appreciated its
content and considered it, like Brown and Uffink, to be more fundamental
than the other laws of thermodynamics.12 Commonly, however, the law is
invoked without being flagged as a law.13 It states:
12See, for instance, Uhlenbeck and Ford (1963: p.5), Kestin (1979: p.72), and Lebowitz(1994: p.135).
13See, for instance, Pauli (1973: p.1) and Sklar (1995: p.20).
21
Minus First Law: An isolated system in an arbitrary initial
state within a finite fixed volume will spontaneously attain a
unique state of equilibrium.
Like the other laws of thermodynamics, the Minus First Law intends to
capture a phenomenological fact. As Brown and Uffink (2001: p.528) explain,
the Minus First Law can be broken into the following three claims:
MFL1. The existence of equilibrium states for isolated systems. The defining
property of such states is that once they are attained, they remain
thereafter constant in time, unless the external conditions are changed.
The claim that such states exist is not trivial—it rules out the possi-
bility of spontaneous fluctuation phenomena.
MFL2. The uniqueness of the equilibrium state; i.e. for any initial state of an
isolated system in a fixed given volume, there is exactly one state of
equilibrium.
MFL3. The spontaneous approach to equilibrium from nonequilibrium. A
nonequilibrium state will typically come about as the result of a re-
moval of internal constraints, such as the rapid displacement of adia-
batic walls separating two bodies.14
Spelled out like this, it’s clear to see that the Minus First Law provides
a characterisation of thermodynamic equilibrium and the approach to equi-
librium. It also reveals that the Minus First Law is asymmetric under time-
reversal. The spontaneous motion towards equilibrium is time-asymmetric
14No indication of the speed of the approach to the new equilibrium state is given ofcourse: thermodynamics provides no equations of motion.
22
because of what equilibrium states, as characterised by MFL1, are: once at-
tained no spontaneous departure from them is possible without intervention
from the environment.
The Minus First Law says that isolated systems will either be in a unique
state of equilibrium or else be spontaneously approaching equilibrium. Com-
pare this to G1. It merely gives this behaviour a name: thermodynamic
behaviour. G1 then, is probably best interpreted as saying that it’s a goal of
nonequilibrium statistical mechanics to provide a microphysical justification
of the Minus First Law. Of course, by providing a microphysical justifica-
tion of the Minus First Law, one can be thought to underpin irreversible
macroscopic behaviour, in line with G2—where irreversible macroscopic be-
haviour is taken to be exhibited by systems that are governed by macroscopic
laws that are time-reversal non-invariant—and one can also be thought to
underpin a particular interpretation of G3; to show how a time-reversal
non-invariant thermodynamics emerges from a time-reversal invariant mi-
crophysics.
It’s worth noting that neither the Minus First Law, the three claims
that constitute it, nor G1 contain adverbs such as “quickly” or “slowly”.
In fact, none of them makes any reference to the rates in which systems
relax to equilibrium. (In fact, none of the goals G1-4 make any reference
to the rates in which systems approach equilibrium.) While it’s sensible to
omit any reference to relaxation rates when articulating the Minus First Law
(since relaxation rates vary significantly from system to system, and because
generality is a feature of a good law) it’s worth bearing in mind, for what’s
to come in the next chapter, that we have detailed information about the
23
rates in which many systems approach equilibrium.
2.1.3 The Zeroth Law
The Zeroth Law, like the Minus First Law of thermodynamics, was invoked
for a long time without notice. While some early writers both noted and
emphasised its importance, it wasn’t formally recognised as a law until 1939,
when Ralph Fowler and Edward Guggenheim coined the expression “Zeroth
Law”.15
We say that two objects are in thermal contact with each other if it’s
possible for heat to flow between them. More will be said about heating
in the upcoming discussion of the First Law of thermodynamics. For now,
it will suffice to say that it’s a form of energy transfer. If we bring two
objects, A and B into thermal contact, then one of three things can happen
as the composite system, subject to the Minus First Law, equilibrates. First,
heat could flow from A to B. Second, heat could flow from B to A. Or
third, there could be no flow of heat between A and B. The Zeroth Law of
thermodynamics says that the third situation is transitive.
Zeroth Law: If A is in thermodynamic equilibrium with B when
they are in thermal contact, and B is in thermodynamic equilib-
rium with C when they are in thermal contact, then A is in
thermodynamic equilibrium with C.
This law allows us to define an important equivalence relation on thermo-
15James Clark Maxwell was an early writer who both acknowledged and emphasised theimportance of what would later become the Zeroth Law of thermodynamics. See Maxwell(1871: p.32-33), for example.
24
dynamic equilibrium states.16 We say that if A and B are thermodynamic
equilibrium states, then they are equitemperature states if and only if A
can be brought into thermal contact with B without the flow of heat.17 The
equitemperature relation is trivially reflexive and symmetric. That it is tran-
sitive is a substantive assumption. The equitemperature relation is needed
to introduce the concept of thermodynamic temperature and to establish a
numerical temperature scale.
The Zeroth Law and, as we’ll see, the First Law of thermodynamics are
not directly related to G1-4—or to any of the reversibility concepts that
were introduced earlier. They are, however, needed to properly understand
versions of the Second Law that are connected to G1-4 and to reversibility
concepts. It is for this reason that they have been included in this presenta-
tion and discussion of classical thermodynamics.
2.1.4 The First Law
Work is a form of energy transfer. One way to increase the internal energy
of a system is to do work on it. For example, when we compress springs
we increase their internal energy by transferring energy to them. That is,
by doing work on them. They retain this energy in the form of potential
energy. This energy can be converted into kinetic energy by allowing them
to do work on their environment. The energy that can be recovered from
16A relation is said to be an equivalence relation, if it is reflexive, symmetric, andtransitive.
17The characterisation of equitemperature is not rich enough to provide us with a nu-merical scale of temperature. In fact, it’s not even rich enough to impose a total orderingon the class of equivalent sets. That is, it cannot be used to make comparative claimsbetween sets of nonequivalent equitemperature states.
25
the compression of an ideal spring that begins in state si and that finishes in
state sf is equal to
W = −∫ sf
si
F · dx. (2.6)
F is the force opposing the direction of compression.
The same amount of energy could, however, be used to stir a viscous fluid.
Suppose, for example, that we attach a compressed ideal spring to a paddle,
immerse it in a viscous fluid, and release it. Interestingly, in this case, we
readily recognise that we’re not able to completely recover as kinetic energy
the energy we put into the system as work. Initially, we might think that
this energy has been lost. But then we notice that the system has gotten
warmer. This suggests, and, in fact, we say, that the energy that went into
the system in the form of work was not lost, but was instead converted into
heat. What makes this suggestion better than the first is that there is a
measurable mechanical equivalent of heat: we can measure the amount of
work that is needed to raise a gram of water by 1 degree, for example.
The First Law of thermodynamics can be thought of as a thermodynamic
expression of the principle of conservation of energy. It says that if an amount
of work W is done on a system, and heat Q passes into it, the internal energy
U of the system is changed by an amount
First Law: ∆U = Q+W.
Sometimes the Law is expressed in differential form.
Differential Form: dU = d Q+ dW.
26
Since U is a state function, we use an exact differential, dU , to represent
change in it. Q and W are not state functions. They are path dependent.
That is why they appear as inexact differentials in the differential form of
the Law.
As we’ve seen, thermodynamics uses the terms “heat” and “work” in
connection with two modes of energy transfer. We can increase the internal
energy of a system either by heating it or by doing work on it. It is assumed,
at least in classical presentations of the theory, that these two modes of
energy transfer can readily be distinguished from one another.
2.1.5 The Second Law
There are several versions of the Second Law of thermodynamics.18 Classical
thermodynamics standardly identifies three. These are: the Kelvin state-
ment, the Clausius statement, and the Entropy statement. This section will
begin by presenting each statement. It will then discuss their relation to one
another and what is typically meant by G4. That is, what authors mean
when they say that it’s a goal of nonequilibrium statistical mechanics to
provide a microphysical justification of the Second Law of thermodynamics.
First, there’s Kelvin’s statement of the Second Law.
Kelvin Statement: No process is possible whose sole result is
the complete conversion of heat into work.19 (Kardar 2007: p.9)
18See Uffink (2001) for an interesting and detailed discussion of the many versions ofthe Second Law of thermodynamics.
19This is a common, modern presentation of Kelvin’s statement of the Second Law. AsUffink (2001: p.327-328) notes, this statement of the Law was inspired by what Kelvinconsidered to be an axiom of thermodynamics:
27
Then there’s Clausius’s statement.
Clausius Statement: Heat can never pass from a colder body to
a warmer body without some other change, connected therewith,
occurring at the same time. (Clausius 1856: p.86)
The first statement rules out a perfect engine. The second rules out a
perfect refrigerator.
Unlike Kelvin and Clausius’s statements of the Second Law, the Entropy
statement cannot be stated and at the same time be so easily understood.
It will help to have some additional concepts on the table, before presenting
this statement of the Law. We begin with Carnot’s theorem.
Carnot’s Theorem
A heat engine works by taking in a certain amount of heat Qh, from a heat
source (e.g. a coal fire), converting a portion of it into work W , and dumping
the remaining heat Qc into a heat sink (e.g. atmosphere).
A heat engine is said to work in a cycle if it returns to its initial thermo-
dynamic state. This is similar to, but distinct from, reversibility understood
as recovery. In this case, auxiliary systems are allowed to be in different
thermodynamic states when the engine returns to its initial thermodynamic
state. A heat engine that operates in a cycle undergoes no net change in
internal energy.
It is impossible, by means of inanimate material agency, to derive mechanicaleffect from any portion of matter by cooling it below the temperature of thecoldest of the surrounding objects. (Kelvin 1851: p.265)
28
Source
Sink
Engine
Qh
Qc
W
Figure 2.1: A heat engine.
The principle of conservation of energy entails that
W = Qh −Qc. (2.7)
The fraction of heat the engine takes from the heat source and converts
into useful work, having dumped the rest, is called the efficiency of the
engine. It’s denoted by η.
η =W
Qh
= 1− Qh
Qc
. (2.8)
A Carnot engine is any heat engine that operates in a cycle in a quasi-
static and reversible manner, and all of its heat exchanges take place either
at a source temperature Th, or at a sink temperature Tc. In these situa-
tions, heat sources and heat sinks are treated as being so large that they can
supply or absorb quantities of heat without incurring a measurable change
in temperature. That is, they are treated as reservoirs. The distinguishing
characteristic of the Carnot engine is that all of its heat exchanges occur at
only two temperatures.
29
Carnot’s theorem gives us information about the maximum efficiency of
heat engines. It has two parts.
Carnot’s theorem: 1. Every Carnot engine that operates be-
tween a pair of heat reservoirs, with temperatures Th and Tc, is
equally efficient. 2. Moreover, any other heat engine operating
between these reservoirs is less efficient than a Carnot engine.
When a Carnot engine runs forward, it draws heat from the hot reservoir
and converts a portion of it into work.
Th
Tc
Carnot Engine
Qh
Qc
W
Figure 2.2: A Carnot engine operates between temperatures Th and Tc,with no other heat exchanges.
Since a Carnot engine is reversible, it can also be run backward and work
as a refrigerator. In this case, work is done on the Carnot engine. This is
used to move heat from the cold reservoir to the hot reservoir (see Figure
2.3).
Carnot’s theorem is proven by deriving a contradiction. Consider a heat
engine, operating in a cycle, with efficiency ηhe, that has been setup to extract
an amount of heat Qh from a hot reservoir, do work W = ηheQh on a Carnot
engine operating in reverse, and discard heat Qc = (1− ηhe)Qh. That is, we
30
Th
Tc
Carnot Engine
Qh
Qc
W
Figure 2.3: A Carnot engine operating in reverse. Heat is transferred fromthe cold reservoir to the hot reservoir.
run the engine forwards, in a cycle, and use it to do work on a Carnot engine
that is extracting heat Q′c from the cold reservoir, and discarding heat Q′h
into the hot reservoir with efficiency ηce.
Th
Heat Engine
Qh
Qc
W Carnot Engine
Q’h
Q’c
Tc
Figure 2.4: A heat engine used to run a Carnot engine in reverse.
The amount of work that is transferred from the heat engine to the Carnot
engine is
W = ηheQh = ηceQ′h. (2.9)
This means that the net effect of the process transfers a quantity of heat
31
Q = Q′h −Qh = (ηheηce− 1)Qh (2.10)
from the cold reservoir to the hot reservoir. If ηhe > ηce, then (2.10) is
positive. The net result of the process, then, is one in which we have moved
heat from the cold reservoir to the hot reservoir. This contradicts Clausius’s
statement of the Second Law.20 So we conclude that for any heat engine and
any Carnot engine, ηhe ≤ ηce. This is the second part of Carnot’s theorem.
The first part follows as a corollary. Every Carnot engine that operates
between a pair of heat reservoirs is equally efficient, since each can be used
to run any other one backward.
Thermodynamic Temperature
Carnot’s theorem says that every Carnot engine operating between a pair
of heat reservoirs has the same (maximum) efficiency. Since their efficiency
depends only on the temperatures of the reservoirs, they can be used to
define a temperature scale. That is, they can be used to define a scale on the
class of sets of equitemperature states that was established using the Zeroth
Law.
If ηhc is the efficiency of a Carnot engine operating between reservoirs H
and C, define the thermodynamic temperature T by
TcTh
=df 1− ηhc. (2.11)
This defines the thermodynamic temperature of any reservoir up to an
20Where Clausius’s statement of the Second Law is taken as an axiom.
32
arbitrary scale factor.
The Carnot Cycle
Functioning Carnot engines perform Carnot cycles. These cycles are usually
divided into four stages. This subsection describes one of these cycles for an
engine working between two reservoirs whose working substance is an ideal
gas.
The first stage involves expansion at constant temperature. This is oth-
erwise know as an isothermal expansion. The engine absorbs heat, Qh, from
the hot reservoir at temperature Th, and does work on the environment. The
second stage involves adiabatic expansion. That is, the gas continues to ex-
pand but now does so without exchanging heat with the environment.21 It
does work on the environment as it cools to temperature Tc. The third stage
involves compression at constant temperature. This is otherwise known as
an isothermal compression. At this stage, an external agent does work on
the gas. The engine expels heat, Qc, into the cold reservoir at temperature
Tc. The fourth and final stage of the cycle involves adiabatic compression.
This stage also involves an agent doing work on the gas, raising its tempera-
ture to Th. The compression occurs without heat being exchanged with the
environment.
21A system is adiabatically isolated if and only if it can’t exchange heat with the envi-ronment. An adiabatic process (e.g. an adiabatic expansion) is one in which the systemexchanges no heat with the environment.
33
Figure 2.5: The four stages of a Carnot Cycle. (A) isothermal expansion.(B) adibatic expansion. (C) isothermal compression. (D) adiabatic
compression. (Schroeder 2000: p.126)
Thermodynamic Entropy
The following equality holds for a Carnot cycle:
∮d Q
T=Qh
Th− Qc
Tc= 0. (2.12)
If we assume that any quasi-static and reversible cycle can be represented
by a collection of Carnot processes, that is, by a path through the system’s
thermodynamic state space that alternates between isothermal and adiabatic
segments, then for any thermodynamic system
∮qsr
d Q
T= 0. (2.13)
The subscript attached to the path integral signifies that the cycle is
performed in a quasi-static and reversible manner. If (2.13) did not hold, then
34
we could construct a quasi-static and reversible heat engine that operated in
a cycle with an efficiency that was different from the efficiency of a Carnot
engine, in violation of Carnot’s theorem.
It follows from (2.13) that there exists a state function S such that, for
any quasi-static and reversible process, and any two thermodynamic states
a and b,
∫ b
a
d Q
T= Sb − Sa. (2.14)
This state function is called the thermodynamic entropy of the system.
It can also be written in differential form.
dS =( d QT
)qsr
(2.15)
For any heat engine operating in a cycle that is less efficient than a Carnot
engine, it must be the case that
∮d Q
T< 0. (2.16)
Equations (2.13) and (2.16), when written together, state that for any
cycle,
∮d Q
T≤ 0. (2.17)
This is known as Clausius’s theorem.22 It’s often expressed in differential
form.
22This is also sometimes referred to as the “Inequality of Clausius”.
35
d Q ≤ TdS. (2.18)
The Entropy statement of the Second Law follows from this result.
Entropy Statement: The thermodynamic entropy of an adia-
batically isolated system cannot decrease. That is, dS ≥ 0.
Relations Between Statements of the Second Law
Three statements of the Second Law have been presented: the Kelvin state-
ment, the Clausius statement, and the Entropy statement. One may wonder
then what could be meant by G4. That is, one may wonder what is meant
by those who say that it is a goal of nonequilibrium statistical mechanics to
provide a microphysical justification of the Second Law of thermodynamics.
Such a claim is likely to have been informed by two thoughts. First, most
authors are aware of the equivalence of these statements. Their equivalence
will be highlighted shortly. Second, it’s reasonable to think that many au-
thors have the Entropy statement in mind when they state this goal. The
reason for this has to do with its content. It, unlike the other statements,
formally identifies a thermodynamic quantity that tracks irreversible macro-
scopic behaviour: thermodynamic entropy. Since (as we’ll see) philosophers
typically have claims such as G1-3 in mind whenever they claim something
like G4, it seems reasonable to interpret them as saying that it’s a goal of
nonequilibrium statistical mechanics to provide a microphysical justification
of why the thermodynamic entropy of adiabatically isolated systems cannot
decrease.
36
Interestingly, one might see the satisfaction of this goal as a way of also
satisfying G2 and G3—where these claims are suitably interpreted. That
is, where we associate increases in thermodynamic entropy with irreversible
macroscopic behaviour (G2), and where we take the Entropy statement of
the Second Law as being a (or perhaps the) time-reversal non-invariant aspect
of thermodynamics that is referred to in G3.
Returning to the equivalence of statements of the Second Law, we find
that the Kelvin and Clausius statements are equivalent if we insist that all
temperatures have the same sign.23 Their equivalence is typically displayed
by highlighting that a violation of one of these statements leads to a violation
of the other, and vice versa.
Consider an engine C that violates the Clausius statement of the Second
Law by transferring Q heat from a colder reservoir to a hotter one. Now
consider a heat engine that operates between these reservoirs that transfers
heat Qh from the hotter reservoir and dumps Qc into the colder one. The
combined system takes Qh − Q from the hot reservoir, produces work equal
to Qh − Qc, and dumps Qc − Q into the colder reservoir. If we adjust the
output of the heat engine such that Qc = Q, the net result is a perfectly
efficient engine K, in violation of Kelvin’s statement. See Figure 2.6.
Now consider an engine K that absorbs Q heat from the hot reservoir
23As Uffink (2001: p.329) notes, Tatyana Ehrenfest-Afanassjewa (1925, 2002) noticedthat the two formulations only become equivalent when we add an extra axiom to ther-modynamics, namely that all temperatures have the same sign. If we allow systems withnegative absolute temperature, which is not forbidden by the standard laws of thermody-namics, then one can distinguish between these two formulations. As Uffink (2001: p.329)continues, this observation became less academic when Norman Ramsey (1956) gave con-crete examples of physical systems with negative absolute temperatures (e.g. nuclear spinsystems.
37
Th
Tc
Q
Q
Qh
Qc= Q
Heat engine C engine W
Th
Tc
Qh- Qc
K engine W =
Figure 2.6: An engine violating the Clausius statement C can be connectedto a heat engine, resulting in a combined device K that violates the Kelvin
statement.
and converts it entirely into work. This engine violates Kelvin’s statement.
The work output by this engine can be used to run a heat engine backwards
(i.e. it can work as a refrigerator), with the net outcome of transferring heat
from a colder body to a hotter body, in violation of Clausius’s statement.
Th
Tc
Q Qh
Qc
Heat engine K engine W
Th
Tc
Qh- Q
C engine =
Figure 2.7: An engine violating the Kelvin statement K can be connectedto a refrigerator, resulting in a violation of the Clausius statement.
Since Kelvin and Clausius’s statements are equivalent, it remains to show
38
that they are equivalent to the Entropy statement. But since Kelvin and
Clausius’s statements are equivalent, we only need to show that one of them is
equivalent to the Entropy statement. And since earlier subsections effectively
detailed how the Entropy statement followed from the Clausius statement,
all that remains is for us to show that the Clausius statement follows from
the Entropy statement.
Consider a violation of the Entropy statement. That is, consider a heat
engine operating in a cycle between two reservoirs with different temperatures
whose thermodynamic entropy decreases, i.e. dS < 0. Then, for such a
process,
d Q > TdS. (2.19)
So
∮d Q
T> 0. (2.20)
This means that we have heat passing from a colder body to a warmer
body without some other change, connected therewith, occurring at the same
time. That is, we have a violation of Clausius’s statement.
2.2 Summary of Results
Through a discussion of various reversibility concepts and the laws of thermo-
dynamics, this chapter highlighted ways in which the goals G1-4 are distinct
but related. This was done so as to make the issue of reconciling the exis-
39
tence of thermodynamically irreversible processes with underlying reversible
dynamics more clear, and to highlight what philosophers are attempting to
underpin using nonequilibrium statistical mechanics. It was also done so that
we might, after the coming section and next chapter, better appreciate the
relationship between thermodynamics and statistical mechanics. This discus-
sion also contributed to the task of explaining what philosophers mean when
they say than an aim of nonequilibrium statistical mechanics is to account
for certain aspects of thermodynamics.
To these ends, an earlier section suggested that G1 is probably best in-
terpreted as saying that a goal of nonequilibrium statistical mechanics is to
provide a microphysical account of the Minus First Law. Earlier sections also
discussed other ways of interpreting this goal and highlighted its connections
to G2-4. It was noted that if one were to provide a microphysical account of
the Minus First Law, one could also be thought to fulfil the goal of accounting
for irreversible macroscopic behaviour, in line with G2—where irreversible
macroscopic behaviour is taken to be exhibited by systems that are governed
by time-reversal non-invariant macroscopic laws—and to satisfy the goal ex-
pressed by a particular interpretation of G3, where the Minus First Law is
thought to be at least one aspect of the theory that is responsible for its
time-reversal non-invariance.
Earlier sections highlighted that G2 could be interpreted in various ways,
depending on how we interpret “irreversible macroscopic behaviour”. It could
be understood as macroscopic behaviour that is governed by some time-
reversal non-invariant law. G2 could also be understood as saying that a
goal of nonequilibrium statistical mechanics is to account for why certain
40
processes render particular initial thermodynamic states irrecoverable. G2
could also be interpreted as a request for a microphysical account of the Minus
First Law, but it could also be interpreted as a request for a microphysical
account of the Second Law of thermodynamics.
How we interpret G3 depends on what aspects of the theory are thought
to be captured by the expression “time-reversal non-invariant thermodynam-
ics”. Again, this could be seen as saying that a goal of nonequilibrium sta-
tistical mechanics is to provide a microphysical account of the Minus First
Law, but it could also be interpreted as saying that a goal of nonequilibrium
statistical mechanics is to provide a microphysical account of the Second Law
of thermodynamics. It could also be understood as saying that it is a goal
of nonequilibrium statistical mechanics to provide a microphysical account
of both laws.
Earlier sections suggested that G4 is probably best interpreted as saying
that it is a goal of nonequilibrium statistical mechanics to provide a micro-
physical account of the Entropy statement of the Second Law of thermody-
namics. This is despite its equivalence to Kelvin and Clausius’s statements
of the law. The reason is that, unlike Kelvin and Clausius’s statements,
the Entropy statement formally identifies a thermodynamic quantity that
tracks irreversible macroscopic behaviour: thermodynamic entropy. It was
also indicated that one might see the satisfaction of this goal as a way of
satisfying G2 and G3—where these goals are suitably interpreted. That is,
respectively, where we associate increases in thermodynamic entropy with ir-
reversible macroscopic behaviour, and where we take the Entropy statement
of the Second Law as being a (or perhaps the) time-reversal non-invariant
41
aspect of thermodynamics that is requested by G3.
2.3 Not Strict Interpretations, Suitable
Analogues
Having highlighted that G1-4 are distinct, and that each goal could be made
more precise—often in a number of ways—one may wonder whether this
points to either a confusion in the literature or a disagreement about what
is central to reconciling the existence of thermodynamically irreversible pro-
cesses with underlying reversible dynamics. One might also think that the
appropriate way for the discussion to proceed is for contributors to first care-
fully specify what it is they’re attempting to use nonequilibrium statistical
mechanics to accomplish.
This chapter will now end by elaborating a little bit on the relationship
between thermodynamics and statistical mechanics, and by explaining why
there is neither confusion nor disagreement in the literature. It will also
explain why well informed authors needn’t worry about, and probably don’t
worry about, eliminating the differences between expressions of goals such as
G1-4.
Those attempting to provide a foundation for thermodynamics recognise
that macroscopic systems can be described in various ways, by various phys-
ical theories. A dilute gas and its behaviour, for example, can be described
in thermodynamic terms. It can also be described at a microscopic level, and
its behaviour, we think, can be modelled by describing the behaviour of the
entities that constitute it. At the microscopic level, a mechanical theory like
42
statistical mechanics is used to describe the system’s behaviour. Since we
think both descriptions are equally valid, the two theories’ predictions had
better be consistent when applied to the same phenomena. Since statisti-
cal mechanics is thought to be the more fundamental theory, the concern is
often transformed into whether we can recover thermodynamics from statis-
tical mechanics. Philosophers often talk about reduction in these contexts.
The worry is otherwise stated in terms of whether one theory reduces to an-
other. This is a way of looking at what those engaged in the foundational
discussion are discussing. While reduction is a philosophically hairy notion,
the main idea shared by many theories of reduction, and implicitly (but
sometimes explicitly) endorsed by those contributing to foundational discus-
sions of thermodynamics, is that one theory reduces to another if we can
use the reducing theory to construct an analogue of the laws and concepts
of the theory to be reduced. What philosophically well-minded contributors
do not expect is a logical deduction from statistical mechanics (the reducing
theory) to thermodynamics (the theory to be reduced). They also do not
expect the laws of the reduced theory to be laws of the reducing theory, nor
do they expect the concepts used by the former to always be applicable at
the level of the latter. The general expectation is instead that the laws and
concepts of thermodynamics appear in statistical mechanics as complicated,
approximate statements that are true under certain conditions.24 To expect
anything more would be, in the words of Craig Callender (2001), to take
thermodynamics too seriously.
24Many authors hold this kind of view. See, for example, Frigg and Werndl (2012a: 919-920), Frigg and Werndl (2012b: p.100), Lavis (2005: p.255), Callender (2001), and Callen-der (1999).
43
With these considerations in mind, it’s reasonable to read goals such as
G1-4 not as ones that are confused, or as ones that are taking a firm stand
on what exactly is needed to reconcile the tension between thermodynam-
ics and statistical mechanics. Instead, it’s reasonable to read them as short
approximate statements that refer to a collection of relevant ideas that are
central to reconciling the existence of thermodynamically irreversible pro-
cesses with underlying reversible dynamics. A collection of relevant ideas,
that is, such as those discussed in previous sections, that come from var-
ious ways of unpacking goals that resemble G1-4. What contributors to
the foundational discussion presumably want from nonequilibrium statistical
mechanics are, among other things, statistical mechanical stories that, at a
macroscopic level, do justice to the aspects of thermodynamics touched on
by claims such as G1-4.
These considerations also motivate why well informed contributors needn’t
bother eliminating the differences between these claims. If they are under-
stood as referring to a collection of important and relevant ideas, and if an
appropriate reduction of thermodynamic concepts to statistical mechanical
ones involves locating suitable analogues, then it’s unnecessary to homogenise
them.
44
Chapter 3
An Evaluation of Foundational
Accounts
The previous chapter discussed one of the aims of statistical mechanics: to
reconcile the existence of thermodynamically irreversible processes with un-
derlying reversible dynamics. It focused on one side of the issue, where it
makes contact with thermodynamics. This chapter takes up another side,
where it makes contact with statistical mechanics.
One of the aims of this chapter is to outline and review some of the leading
attempts to reconcile this conflict. These attempts are typically part and
parcel of works that aim at accounting for why isolated systems that begin
away from equilibrium spontaneously approach equilibrium and why they
remain in equilibrium for incredibly long periods of time. In addressing this
goal, these works hope to provide a foundation for aspects of thermodynamics
and satisfy goals such as G1-4. This chapter considers the most pressing
problems facing these accounts and, where possible, presents and discusses
45
their most promising solutions. It highlights that they have some severe
limitations, even when we overlook their problems. It reveals that they do
not underpin a large and important set of facts that concern the behaviour
of systems that begin away from equilibrium. While they may satisfy goals
such as G1-4 and account for why isolated systems that begin away from
equilibrium spontaneously approach equilibrium, and why they remain in
equilibrium for incredibly long periods of time, they do not underpin facts
about the rates in which systems approach equilibrium, or facts about the
kinds of states they pass through on their way to equilibrium, or facts about
fluctuation phenomena. They also do not help us form expectations about
these things or help justify the expectations we may already have about them,
having formed them on the basis of experience.
This chapter claims that the limitations of these accounts are a symptom
of what they are aiming at. By focusing on recovering certain aspects of
thermodynamics, those contributing to the foundational project have merely
been in the business of articulating microphysical accounts of irreversible
macroscopic behaviour that recover a few qualitative facts. To remedy this
situation, this chapter encourages pursuing a different line of research: un-
derstanding and accounting for the success of the techniques and equations
physicists use to model the behaviour of systems that begin away from equi-
librium. If we’re able to properly understand these techniques and equations,
and are able to account for their success, then we will not only be able to
underpin the kinds of qualitative facts the literature has focused on, but
we’ll also be able to underpin the important quantitative facts that leading
typicality accounts cannot.
46
The following chapter takes steps in this promising direction. It outlines,
discusses, and attempts to ground the success of a way of modelling the
approach to equilibrium and subsequent behaviour of a Brownian particle
that’s been introduced into an isolated homogeneous fluid at equilibrium.
The next chapter also highlights how the model contributes to the satisfaction
of goals such as G1-4, and to the underpinning of important quantitative
facts.
This chapter is composed of five sections. The first section outlines and
discusses the essence of what many people think of as the first serious at-
tempt to account for the behaviour of systems that begin away from equi-
librium. This approach is based on some of Ludwig Boltzmann’s early work
on statistical mechanics, when he first offers what has come to be known as
Boltzmann’s equation and H-theorem. As we’ll see, this work gives rise to
one of the very issues that is at the centre of contemporary discussions of
statistical mechanics: the reconciliation of the existence of thermodynami-
cally irreversible processes with underlying reversible dynamics. The second
section outlines and discusses the leading ways—found in the contemporary
foundational literature—to account for the behaviour of systems that be-
gin away from equilibrium. These approaches are inspired by, and build
on, Boltzmann’s later attempts (circa 1877) to resolve this issue and, once
again, account for the behaviour of systems that begin away from equilib-
rium.1 Each one incorporates, in some way or other, a notion of typicality.
1This presentation and discussion of Boltzmann’s work is couched in the language ofmodern statistical mechanics. It also makes use of its technical resources. While thispresentation does not do justice to the messy and tangled history of statistical mechanics,or to Boltzmann’s original works, it is in line with how these ideas are now standardlypresented. It also happens to be the cleanest way to present the ideas most relevant to
47
The second section also discusses worries specific to each typicality account.
More general concerns are discussed in the third section. Both sections also
discuss, where possible, the most promising solutions to these concerns. The
fourth section highlights and discusses the limitations of typicality accounts.
It also promotes pursuing lines of research that aim at more than merely
recovering a few qualitative facts and satisfying goals such as G1-4. The
suggestion is that we make sense of, and try to ground the success of, the
techniques and equations commonly used to model the behaviour of systems
that begin away from equilibrium. Understanding and accounting for their
success will provide us with a microphysical underpinning of the aspects of
the behaviour of systems that begin away from equilibrium not accounted for
by typicality accounts. It will also provide us with the means to satisfy goals
such as G1-4. The chapter ends, in the fifth section, with a brief summary.
3.1 Boltzmann’s Early Work
Ludwig Boltzmann famously attempted to explain why isolated systems that
begin in some nonequilibrium state spontaneously approach equilibrium and
why they remain in equilibrium for incredibly long periods of time. In 1872,
Boltzmann considered how the distribution of velocities of the molecules of a
dilute gas could be expected to change under collisions and argued that there
the aims of this chapter and thesis. Readers interested in the gritty details are encouragedto consult Klein (1973), Brush (1986), Sklar (1995), von Plato (1994), Cercignani (1998),Ehrenfest and Ehrenfest (2002), Uffink (2007), Brown, Myrvold, and Uffink (2009), andUffink (2014) for discussions of Boltzmann’s work on statistical mechanics and its tangledhistory. For Boltzmann’s original works, see Boltzmann (1909). And see de Regt (1996),Blackmore (1999), and Visser (1999) for discussions of the philosophical and methodolog-ical shifts in Boltzmann’s thinking.
48
was a unique distribution—now called the Maxwell-Boltzmann distribution—
that was stable under collisions.2
Maxwell-Boltzmann Distribution: the fraction of molecules
having velocity lying within the velocity-space element d3v about
v is proportional to
e−mv2/2kT d3v, (3.1)
where k is the Boltzmann constant and T is temperature.
Boltzmann further argued that a gas that initially had a different distribu-
tion would move toward the Maxwell-Boltzmann distribution. To argue that
this was the unique (equilibrium) distribution, which would be approached
starting from some other distribution, Boltzmann defined a quantity, which
we now call H, showed that it reached a minimum value for the Maxwell-
Boltzmann distribution, and argued that it would monotonically decrease to
its minimum.3 This result is now known as Boltzmann’s H-theorem. It’s
a straightforward consequence of Boltzmann’s transport equation. Impor-
tantly, the result is asymmetric under time-reversal.
The next subsection fleshes this picture out by presenting and discussing
aspects of the theorem. The subsection after that highlights its logic and
limitations using a simple toy model that shares salient features with dilute
gases.4
2See Boltzmann (1872).3The quantity we call H was originally denoted E in Boltzmann’s early work. See
Boltzmann (1872).4The interested reader should consult Uffink (2007) and, especially, Brown et al. (2009)
for the details of Boltzmann’s H-theorem, and for an in depth discussion of it. For a good
49
3.1.1 Boltzmann’s H-theorem
Boltzmann’s H-theorem was based on a model of a gas consisting of n qual-
itatively identical, hard, spherical molecules, in a container with perfectly
elastic walls. He supposed the gas was sufficiently dilute so that only binary
collisions needed to be taken into account in the dynamics.
The distribution (or density) function f(x,v, t) is defined such that
f(x,v, t)d3xd3v is the number of molecules within a volume element d3x
about x and with velocity lying within the velocity-space element d3v about
v. Boltzmann’s transport equation determines how f(x,v, t) evolves in time.
Brown et al. (2009: p.175) explain that Boltzmann’s first derivation of
the transport equation made use of three noteworthy assumptions. Two
of these were harmless, the third became controversial. First, Boltzmann
assumed that for the initial state of the gas, the momentum distribution is
isotropic. That is, f0(x,v, t) = f0(x, v, t). Second, Boltzmann assumed that
no external forces act on the gas and that the distribution is independent
of position at all times. That is, f(x,v, t) = f(v, t). It follows from these
considerations that the expression for ∂f/∂t depends only on collisions, so
that the transport equation can be expressed as a balance equation, in which
losses are subtracted from gains during collisions. The third and most famous
assumption concerned collision numbers. We now refer to this assumption,
following Paul and Tatiana Ehrenfest, as the Stoßzahlansatz.5 Here’s how
Brown et al. (2009: p.175) describe it:
heuristic derivation of Boltzmann’s H-theorem see Dorfman (1999).5See Uffink (2007) and Brown et al. (2009) for more on the Stoßzahlansatz.
50
Suppose we have two groups of molecules heading towards each
other. Choose one of the groups and call it the target group.
Consider the collection S of all the spatial volumes (cylinders)
swept out by the target molecules in time δt such that if any
molecule of the second group is found in S there is bound to be
a collision. Note that S depends on the relative velocities of the
molecules in the two groups; it is not defined solely in relation to
the target molecules! Then suppose that the number of collisions
that actually occur is the total volume associated with S multi-
plied by the density of molecules of the second, colliding group.
This holds only if the spatial density of the colliding molecules in
S is the same as in any other part of space.
Since the target region S is defined in relation to the relative velocities of
the molecules from the two groups, and since, from the second assumption,
the distribution function is position independent, the assumption depends on
the relationship between the momenta of the target and colliding molecules
before collision. In fact, the Stoßzahlansatz is tantamount to the claim that
the density F (v1, v2, t) of pairs of molecules which are about to collide within
the interval [t, t+ ∆t] with velocities v1 and v2 is given by the product of the
single molecule densities. That is,
F (v1, v2, t) = f(v1, t)f(v2, t). (3.2)
Thus the molecules entering into but not out of collision are uncorrelated
in velocity.
51
In the same work, Boltzmann also introduced the H-functional
H[ft] =
∫f(x,v, t) ln f(x,v, t)d3xd3v (3.3)
which is large for narrow distributions and small for wide ones. Boltz-
mann treated the density function as continuous and differentiable, since the
number of molecules in a gas is very large.
If we assume that the Stoßzahlansatz holds at time t0, then it follows
from Boltzmann’s transport equation that
dH
dT≤ 0. (3.4)
Equality holds when the gas reaches equilibrium, coinciding with the
Maxwell-Boltzmann distribution. This is Boltzmann’s H-theorem. One of
its assumptions is the validity of the Stoßzahlansatz at all times. This en-
sures the monotonic behaviour of H over time. It also ensures that it’s a
temporally asymmetric result. The theorem can be thought to provide a
microphysical underpinning of the Minus First Law. It can also be thought
to provide an underpinning of the Entropy Statement of the Second Law of
thermodynamics, if we associate thermodynamic entropy with −H. It can
also be thought to satisfy G2.
3.1.2 The Kac Ring
This subsection highlights the theorem’s logic and limitations by discussing
a simple, explicitly solvable toy model called the Kac ring. The Kac ring
first appeared in a series of lectures given by Mark Kac at the University of
52
Colorado in 1959.6 It’s similar, in some important respects, to a dilute gas.
In the model, N sites are arranged around a circle, forming a one-
dimensional periodic lattice. Sites are joined to their neighbours by an edge,
and 0 < n < N of the edges carry a marker. Each site is occupied by either a
black ball or a white ball.7 The balls and markers are similar to the molecules
that comprise a gas.
Figure 3.1: A Kac ring with N = 8 lattice sites and n = 5 markers.
The system evolves on a discrete set of ticks t ∈ Z from state t to state
t + 1 in the following way: each ball moves in a clockwise direction to its
6The purpose of these lectures was to furnish an introduction to probability theory andits applications to an audience that had little knowledge of these subjects. In a lectureon classical statistical mechanics, Kac (1959: p.99) used the ring model to introduce hisaudience to the statistical mechanical treatment of irreversible phenomena. The Kac ringis often used to introduce this idea. It’s also used to introduce some related ideas. See,for example, Bricmont (1995: Appendix 1), Bricmont (2001), Dorfman (1999: Sec.2.3),Kac (1959: Ch.3 Sec.14-15), Gottwald and Oliver (2009), Schulman (1997: Sec.2.1), andThompson (1972: Sec.1.9).
7Similar descriptions of the model can be found in Bricmont (1995: Appendix 1), Bric-mont (2001: p.10), Dorfman (1999: Sec.2.3), Kac (1959: p.99), and Gottwald and Oliver(2009: Sec.3).
53
nearest neighbour. When a ball passes a marker its colour changes. This is
analogous to changes in the velocities of molecules of a gas as they collide
with one another.
The Kac ring has a number of interesting features. Analogous to Boltz-
mann’s dilute gas, its microdynamics are symmetric under time-reversal.
Any reversed sequence of states is compatible with the dynamics of the sys-
tem. The system is strictly periodic, and so displays recurrence. The ring,
after a series of ticks, returns to its initial state. After N ticks, each ball has
reached its initial site and changed colour n times. If n is even, the initial
state recurs. If n is odd, it takes at most 2N ticks for the initial state to
recur. Recurrence is a property also shared by contained gases.
Let B(t) denote the total number of black balls and b(t) the number of
black balls that pass a marker on the next tick. Similarly, let W (t) denote
the number of white balls and w(t) the number of white balls that pass a
marker on the next tick. It follows that
B(t+ 1) = B(t) + w(t)− b(t) (3.5)
and
W (t+ 1) = W (t) + b(t)− w(t). (3.6)
We can study the difference between the number of black and white balls
at various times.
∆(t) = W (t)−B(t) (3.7)
54
and
∆(t+ 1) = W (t+ 1)−B(t+ 1) = ∆(t) + 2b(t)− 2w(t). (3.8)
The system is said to be in equilibrium when W (t) ≈ B(t). W , B, and
∆ are macroscopic quantities. They describe global features of the system’s
state; many different microstates give rise to the same macroscopic quantities.
In contrast, w and b give local information about individual sites. They
cannot be computed without knowing the location of each marker and the
colour of the ball at each site. Importantly, the evolution of W , B, and ∆
cannot be determined using only macroscopic state information.
This limitation, however, can be overcome if we make the following non-
dynamical assumption: suppose that the fraction of white or black balls that
change colour at each tick is equal to the probability µ that an edge has a
marker on it, where µ is equal to the number of markers, n, divided by the
number of edges, N . That is,
µ =n
N=
w(t)
W (t)=
b(t)
B(t). (3.9)
This assumption is analogous to the assumption that the Stoßzahlansatz
holds at all times. We too have posited a continued independence. Here, the
colour of each ball is taken, at each tick, to be probabilistically independent
of whether there is a marker in front of it. It’s important to note that we
have also introduced a time-reversal non-invariant element into the model.
There are sequences of states of the system that are compatible with the
assumption whose temporal reverse is not. For example, consider a ring that
55
has a white ball at each site and whose markers at some initial time have been
randomly distributed. Now let the system evolve for one tick. All and only
the balls that have changed colour have passed a marker. The assumption,
which holds for this sequence of states, does not hold for its time reverse. In
the later case, ball colours are not independent of marker locations. Black
balls are found at all and only those sites that have a marker in front of them.
Importantly, this collision assumption enables us to express (3.8) as
∆(t+ 1) = ∆(t) + 2µB(t)− 2µW (t) = (1− 2µ)∆(t). (3.10)
This yields
∆(t) = (1− 2µ)t∆(0), (3.11)
which is analogous to Boltzmann’s equation. Since 0 < µ < 1, (3.11)
tells us to expect |∆(t)| → 0 as t→∞. This is the analogue of Boltzmann’s
H-theorem. This result, like Boltzmann’s H-theorem, is asymmetric under
time-reversal. Moreover, this result, like Boltzmann’s, is inconsistent with
the system’s dynamics. More will be said about this in a moment.
Returning to the H-theorem, we may ask how Boltzmann arrived at his
result having only assumed an underlying dynamics that is symmetric under
time-reversal. The answer is that he didn’t, and two famous objections have
shown that he couldn’t. These are known as the reversibility objection and
the recurrence objection. The former is usually credited to Josef Loschmidt
and the latter to Ernst Zermelo.8
8See Uffink (2007) and Brown et al. (2009) for more on these objections.
56
The reversibility objection notes that for any set of trajectories of the
molecules of a gas, the time-reversed trajectories are also possible. This is a
straightforward consequence of assuming that the microdynamics is symmet-
ric under time-reversal. It shows that not all microstates of the gas at any
time lead to a monotonic decrease of H. The recurrence objection applies
to classical systems with bounded phase spaces. That is, to systems, such
as Boltzmann’s dilute gas, with total fixed energy. If we consider a small
open neighbourhood of the initial state, and ask, will the state of the system,
after it leaves that neighbourhood, ever return to it? Then the answer, which
makes use of Henri Poincare’s recurrence theorem, is yes, it will, for almost
all initial phase space-points, i.e. for all except a set of Lebesgue measure
zero.
The same objections apply to the Kac ring. Its underlying dynamics are
symmetric under time-reversal, so not all microstates of the system compati-
ble with its macroscopic properties at any time lead to a monotonic decrease
of |∆(t)|. This is an instance of the reversibility objection. Moreover, because
the system is strictly periodic, no initial microstate will yield a monotonic
decrease of |∆(t)|. This is an instance of the recurrence objection.
To derive Boltzmann’s original, asymmetric, result, one needs more than
what is given by simply applying Newton’s laws of motion to molecular col-
lisions. For Boltzmann, it was the assumption that the Stoßzahlansatz holds
at all times. This is analogous to assuming that (3.9) holds at each tick when
deriving (3.11), a monotonic and temporally asymmetric result.
In the wake of concerns about Boltzmann’s H-theorem, many began to
wonder how irreversible thermodynamic behaviour could emerge from an
57
underlying reversible dynamics. That is, one of the very issues that is at
the centre of philosophical discussions of statistical mechanics emerged from
worries about Boltzmann’s original attempt to account for the behaviour of
systems that begin away from equilibrium. These concerns have also influ-
enced many to regard the Second Law of thermodynamics as an approximate
statement that holds with high probability. These concerns have also influ-
enced many to endorse a weakened version of it, and to seek justification for
a revised statistical version of Boltzmann’s H-theorem, along the lines that,
for macroscopic systems, H will probably decrease to a minimum, and stay
there for long periods of time. Most physicists now endorse something like
the following statement.
Revised Second Law: Although fluctuations will occasionally
result in heat passing spontaneously from a colder body to a
warmer body, these fluctuations are inherently unpredictable and
it is impossible for there to be a process that consistently and
reliably harnesses these fluctuations to do work.
3.2 Typicality Accounts
In the years that have followed Boltzmann’s derivation of the H-theorem,
a number of authors have attempted to reconcile the existence of thermo-
dynamically irreversible behaviour with an underlying reversible dynamics.
They have offered accounts that aim at satisfying goals such as G1-4. More
directly, however, these accounts aim at explaining why isolated systems
spontaneously approach equilibrium and why they remain in equilibrium
58
for incredibly long periods of time. An interesting and important class of
these accounts have been inspired by, and build on, Boltzmann’s later at-
tempts (circa 1877) to justify the H-theorem. Each of these accounts takes
the lessons learnt from the reversibility and recurrence objections seriously.
They attempt to show that irreversible macroscopic behaviour is, in some
sense, typical, and they often try to recover a revised version of the Second
Law of thermodynamics. Intuitively speaking, something is typical if it hap-
pens in the “vast majority” of cases. Typical lottery tickets, for example,
lose. Typicality accounts usually focus on accounting for the behaviour of
Boltzmann-like gases. More often than not, though, they are thought to be
applicable to a wide variety of systems.
Standardly, authors claim that Boltzmann (1877) inspired the modern
typicality approach. Joel Lebowitz (1993a, b) and Shelly Goldstein (2001)
are often credited as having ushered in modern forms of the view.9 Typicality
views have become increasingly popular, since the appearance of their work.
This section outlines several of the most promising and influential typ-
icality accounts.10,11 It also presents and discusses a number of their most
9See also, Lebowitz (1999) and Goldstein and Lebowitz (2004). Jean Bricmont (1995)and Roger Penrose (1989) are sometimes, but much less often, also credited with estab-lishing modern forms of the view.
10To be fair, some of the accounts presented in this section do not trade under thename “typicality account”. At least not by those endorsing them. Both the ergodicview and the epsilon ergodic view fall under this category. Roman Frigg and CharlotteWerndl, for example, have appealed to epsilon ergodicity to explain why certain systemsspontaneously approach equilibrium and why remain in equilibrium for long periods oftime. (See Frigg and Werndl 2011, and Frigg and Werndl 2012b.) Interestingly, they bothexplicitly distance themselves from typicality when they articulate their view. As thissection highlights, however, both (what will be called) the ergodic view and the epsilonergodic view can straightforwardly be read through a typicality lens. Since these viewsare subject to the same limitations standard typicality accounts face, it is both fitting andconvenient to present them as a kind of typicality account.
11A less promising and less influential typicality account that is not discussed is con-
59
worrying concerns. It also discusses their most promising solutions. The
next section continues this discussion by focusing on more general, but no
less worrying, concerns with typicality accounts. It also discusses their most
promising solutions. A later section highlights the limitations of typicality
accounts. It notes, in particular, that even if we overlook the most pressing
concerns with typicality accounts, that these accounts will not have achieved
very much. Their shortfalls will be used to motivate a different line of re-
search: to encourage philosophers to consider, understand, and account for
the success of the techniques and equations physicists use to model the be-
haviour of systems that begin away from equilibrium.
3.2.1 Some Background
Consider, once again, Boltzmann’s dilute gas. It’s an isolated system, con-
sisting of n classical particles, each with three degrees of freedom, that are
confined to a container with perfectly elastic walls. The system has a fixed
total energy, E. The microstate of the system is specified by a point x, in
6n-dimensional phase space, Γ. Γ is endowed with Lebesgue measure µL.
The system is governed by Hamilton’s equations of motion, which define a
measure preserving flow φt on Γ. This means that for every measurable set
A, and every time t, µ(φ−1t (A)) = µ(A). The system’s initial microstate at
nected to the view that systems are what Frigg (2009: p.84) calls “globally entropy in-creasing”. For more on this view see Frigg (2009: pp.84-87), and the references therein.Other less promising and less influential typicality accounts that are not discussed in thisthesis, or rather, accounts that could straightforwardly be offered within the typicalityframework but that will not be discussed in this thesis, include the view that relevantsystems are mixing and the view that relevant systems have an ergodic decomposition.For work related to these views, see Frigg and Werndl (2012b: Sec.4.3). See also Sklar(1995: pp.53-59), Bricmont (2001: p.16), and Lavis (2008: Sec.2).
60
time t0, x(t0), evolves into x(t) = φtx(t0) at time t. For isolated Hamiltonian
systems, such as Boltzmann’s gas, energy is conserved. So the motion of
the system is confined to its 6n-1-dimensional energy hypersurface ΓE. The
measure µL can be restricted to ΓE, inducing an invariant measure µ on ΓE.
Thermodynamic systems, as noted in the previous chapter, are charac-
terised by a set of macrovariables: volume, temperature, pressure, etc.12
Macrostates of the system correspond with the macrovariables obtaining cer-
tain values, or with some range of values, if we take into consideration our
means of distinguishing between them.13 Each macrostate Mi, i = 1, 2, . . . ,m
(where m is finite) corresponds to a set of macro-regions ΓMithat consist of
all x ∈ Γ that take the macrovariable values characteristic of Mi. Together,
the ΓMiform a partition of ΓE. That is, they do not overlap but jointly
cover ΓE. Symbolically, ΓMi∩ ΓMj
= ∅ for all i 6= j and i, j = 1, . . . ,m, and
ΓMi∪ . . . ∪ ΓMm = ΓE.14
Mp is the initial macrostate of the system and Meq is the equilibrium
macrostate. Meq is characterised as the macrostate whose associated macrovari-
ables remain (approximately) constant in time.15 For a Boltzmann-like gas,
Meq is standardly identified as the macrostate whose temperature, pressure,
12There are typically a variety of sets one can use to characterise a thermodynamicsystem. Physicists often choose sets based on the problems they’re dealing with.
13It’s standardly assumed that each macrostate compatible with the constraints placedon the system can be characterised using macrovariables that take on well defined values,or that are at least compatible with some well defined range of values.
14As usual, ‘∩’, ‘∪’, and ‘∅’ denote set theoretic union, intersection, and the empty set,respectively.
15See Werndl and Frigg (2014) for an interesting discussion of the consequences of thischaracterisation of equilibrium when viewed through the lens of Boltzmannian statisticalmechanics; they prove a new theorem establishing that equilibrium thus defined corre-sponds to the largest macroregion of a system’s phase space energy hypersurface. Also,notice the similarity of this characterisation of equilibrium with the thermodynamic char-acterisation outlined by Brown and Uffink (2001).
61
and volume remain (approximately) constant through time. Corresponding
to these macrostates are, respectively, the macro-regions ΓMp and ΓMeq .
The Boltzmann entropy of a macrostate Mi is standardly defined as
SB(Mi) =df k log[µ(ΓMi)], where k is again Boltzmann’s constant. Since
a system is in exactly one macrostate at a time, we define the Boltzmann en-
tropy of a system at a time t, SB(t), as the entropy of the system’s macrostate
at t. That is, SB(t) =df SB(Mx(t)) where x(t) is the system’s microstate at t
and Mx(t) is its corresponding macrostate.
Typicality views attempt to account for why systems beginning in Mp end
up in Meq, and why they remain in Meq for incredibly long periods of time.
They also attempt to show that SB(t) generally increases to its maximum
value as the system evolves from Mp to Meq and that it remains (approxi-
mately) at this value for incredibly long periods of time. The later is done
so as to claim that SB(t) is the microphysical analogue of thermodynamic
entropy.
3.2.2 The Dominance View
One kind of typicality account, which we’ll call the dominance view, at-
tempts to account for the approach to equilibrium and an associated en-
tropy increase by arguing for them on the basis of a technical result that
applies to Boltzmann-like gases known as the “dominance of the equilibrium
macrostate”.
The first ingredient in the account is an interpretation of the measure over
microstates on ΓE—standardly taken to be Lebesgue measure—as a typical-
ity measure. Typicality measures represent the relative size of sets of states.
62
They are usually understood to be a kind of normalised measure. Typical
states show a certain property if the measure of the set that corresponds to
this property is one or close to one. What it means to be a typicality measure
is one of the pressing general concerns that will be discussed in a later section.
The second, closely related ingredient is the idea that Lebesgue measure is
a good (perhaps, the appropriate) measure of what’s typical. That is, that
it’s a good typicality measure. Justifying why Lebesgue measure is a good
choice of typicality measure is another pressing general concern that will be
discussed in a later section.
The next ingredient is the recognition that under certain circumstances,
and for Boltzmann-like gases, Meq is the largest of all Mi.16 In fact, for
large n, Meq is enormously larger than any other region.17 For example, the
ratio µ(Meq)/µ(Mi) is of the magnitude of 10n, where Mi is a nonequilibrium
macrostate of an ordinarily prepared system.18
Since, for large n, ΓE is almost entirely taken up by equilibrium mi-
crostates, the dominance of the equilibrium macrostate is often taken to im-
ply that equilibrium microstates are typical with respect to ΓE and Lebesgue
measure µ.19 Some authors claim that its dominance explains why systems
exhibit the following qualitative facts: that isolated macroscopic systems that
begin away from equilibrium spontaneously approach equilibrium and that
16See Frigg (2011: pp.89-90) and Uffink (2007: pp.974-983) for a discussion of the cir-cumstances in which this result holds. See Werndl and Frigg (2015) for an argument thatattempts to generalise this result.
17See Ehrenfest and Ehrenfest (2002: p.30).18See Frigg (2011: p.80), Goldstein (2001: p.43), and Penrose (1989: p.403).19See, for example, Frigg (2011: p.81), Bricmont (2001: p.146), Goldstein (2001: p.43),
and Zanghı (2005: p.191 & p.196).
63
Meq
Figure 3.2: The dominance of the equilibrium macrostate on the energyhypersurface.
they stay there for incredibly long periods of time.20 Some authors also claim
that its dominance explains why thermodynamic entropy generally increases.
Naturally, this account is seen by some as a way of fulfilling goals such as
G1-4 without introducing a time-reversal non-invariant element.
Frigg (2011: p.81) claims that such a view may be read into the following
passage:21
. . . reaching the equilibrium distribution in the course of the tem-
20See Frigg (2011: Sec.4.3) for a list of authors that endorse this kind of view.21Frigg (2011: p.81) also claims that the dominance view may be read into the following
passage by Goldstein, but this quote is taken from an article that is standardly thoughtto endorse what the next subsection labels “the unspecified dynamical view”.
Suppose a system, e.g. a gas in a box, is in a state of low entropy at sometime. Why should its entropy tend to be larger at a later time? The rea-son is basically that states of large entropy correspond to regions in phasespace of enormously greater volume than those of lower entropy. (Goldstein2001: p.49)
64
poral evolution of a system is inevitable due to the fact that the
overwhelming majority of microstates in the phase space have
this distribution; a fact often not understood by the critics of
Boltzmann. . . (Zanghı 2005: p.196; Frigg’s translation)
The view can also be found in Price (1996: pp.39-40).
Thus it seems to me that the problem of explaining why entropy
increases has been vastly overrated. The statistical considerations
suggest that a future in which entropy reaches its maximum is
not in need of explanation; and yet that future, taken together
with the low-entropy past, accounts for the general gradient.
The most troubling concern for the dominance view is its lack of a con-
nection to the dynamics.22 Nothing that has been said so far guarantees,
or even makes it likely, that the microstates of systems prepared into some
nonequilibrium macrostate evolve into states that constitute the equilibrium
macrostate. It also doesn’t follow from the view presented above that systems
whose microstates do evolve into the equilibrium macrostate remain in it for
even short periods of time. Or even that states that begin in equilibrium stay
there! The idea that systems remain in states away from equilibrium is per-
fectly consistent with the idea that equilibrium microstates are typical. So
too is the idea that they fluctuate wildly in and out of equilibrium. The ques-
tion then of what kinds of dynamical considerations need to be added to the
view, along with any restrictions that need to be placed on the kinds of initial
microstates we allow—on the basis of available preparation procedures—so
22Other concerns are discussed in Frigg (2009) and Frigg (2011). See also Uffink (2014).
65
that it can underpin the sought after qualitative facts, and satisfy goals such
as G1-4, leads us onto discussing other typicality accounts.
3.2.3 The Unspecified Dynamical View
Aware of concerns with the dominance view—including, most troublingly,
the concern mentioned above, that measure theoretic considerations cannot,
on their own, tell us anything about the behaviour of systems that begin
away from equilibrium—many supporters of typicality incorporate into their
view some claim about a system’s dynamics and initial state.23 Standardly,
they claim that given any reasonable account of the dynamics, typical initial
microstates are taken by the dynamics into the equilibrium macrostate and
that they remain there for incredibly long periods of time. Let’s call this the
unspecified dynamical view. The classic and often cited expression of this
view is found in Goldstein (2001: pp.43-44):
[ΓE] consists almost entirely of phase points in the equilibrium
macrostate [Meq], with ridiculously few exceptions whose total-
ity has volume of order 10−1020 relative to that of [ΓE]. For a
non-equilibrium phase point [x] of energy E, the Hamiltonian dy-
namics governing the motion [x(t)] would have to be ridiculously
special to avoid reasonably quickly carrying [x(t)] into [Meq] and
keeping it there for an extremely long time—unless, of course, [x]
itself were ridiculously special.24
23For a list and discussion of other concerns, see Frigg (2009) and Frigg (2011). See alsoUffink (2014).
24This quote has been modified to be consistent with the formalism that has alreadybeen introduced. Its content has not been altered.
66
The unspecified dynamical view is, at present, the most promising and
influential kind of typicality view. One of its ingredients is, like before, an in-
terpretation of the measure over microstates on ΓE as a typicality measure.
Another is that it’s a good choice of typicality measure. But here, unlike
the dominance view, which treats the relevant typicality property as being
a microstate that corresponds to the equilibrium macrostate, the relevant
property is (or is something close to) being a microstate that has been pre-
pared in an ordinary way, that evolves under the dynamics into a microstate
that corresponds to the equilibrium macrostate (if it wasn’t already in that
macrostate), and that remains in the equilibrium macrostate for long periods
of time.
When supporters of this kind of view discuss the dynamics of relevant
systems, it’s common for them to write very general and often noncommittal
things. The description offered above by Goldstein is a good example. Here’s
another example, offered by Frigg (2008b: p.114):25
If we now assume that the system’s state drifts around more or
less “randomly” on [ΓE] then, because [Meq] is vastly larger than
any other macro region, sooner or later the system will reach
equilibrium and stay there for at least a very long time.
As Frigg (2008b: p.114) rightly notes, the qualification “more or less ran-
domly” is essential. If its motion were just right, then it could avoid ever
wandering into the equilibrium macrostate. Or else, it could, among other
things, wander in and out of this state in a very erratic fashion, thereby
giving rise to a kind of anti-thermodynamic behaviour.
25See Lazarovici and Reichert (2014: p.7), also.
67
While some are dissatisfied by these descriptions, and think that a more
detailed account of the dynamics is necessary, supporters of this view often
see the generality of their position as both reasonable and virtuous.26 Their
point is often that there are many different dynamical properties a system
could possess that each have the consequence that thermodynamic behaviour
is typical, given the dominance of the equilibrium macrostate. And that the
dynamics would need to be very precise indeed to avoid this consequence,
since we’re unable to prepare systems into states that would otherwise yield
anti-thermodynamic behaviour.
Those unsatisfied with the unspecified dynamical view often try to supple-
ment it by claiming that relevant systems possess some dynamical property
found on the ergodic hierarchy.27 Properties standardly appealed to include:
ergodicity and epsilon-ergodicity.28
26Uffink (2007), Frigg (2009), Frigg (2011), Frigg and Werndl (2011), and Uffink (2014)express dissatisfaction at the manner in which generality is obtained by this view. Friggand Werndl (2012b: p.101), for example, say:
Just saying the relevant Hamiltonians possess the dynamical property of TD-likeness has no explanatory power—it would be a pseudo-explanation of thevis dormitiva variety. The challenge is to identify in a non-question beggingway a dynamical property (or, indeed, properties) that those Hamiltonianswhose flow is TD-like have.
Lazarovici and Reichert (2014: p.31-34), for example, see the generality of the position asboth reasonable and virtuous.
27See Frigg, Berkovitz, and Kronz (2014) and Berkovitz, Frigg, and Kronz (2006) formore on the ergodic hierarchy.
28Some authors appeal to some form of mixing, but very few, if any, claim that relevantsystems are K-systems or Bernoulli.
68
3.2.4 The Ergodic View
While no one contributing to the contemporary discussion appears to endorse
what we’ll call the ergodic view, it both frequently appears in contemporary
discussions and is regarded by many to capture a position carved out by
some of the ideas Boltzmann expressed in his later writings on statistical
mechanics. Understanding this view makes it easier to understand a closely
related view that is taken seriously, so it’s worth discussing here.
Boltzmann conjectured that
The great irregularity of the thermal motion and the multitude
of forces that act on a body make it probable that its atoms,
due to the motion that we call heat, traverse all positions and
velocities which are compatible with the principle of [conservation
of] energy. (Quoted in Uffink (2007: p.40))
This has come to be known as the ergodic hypothesis. As stated, it cannot
be correct, as the trajectory is a one-dimensional continuous curve and so
cannot fill a space of more than one dimension. But it can be true that
almost all trajectories eventually enter every open neighbourhood of every
point on the energy surface.29
A dynamical system 〈Γ,S, µ, φt〉 with a measure preserving evolution (i.e.
for every measurable set A, and every time t, µ(φ−1t (A)) = µ(A)) is said to
29Boltzmann argued, on the basis of the ergodic hypothesis, that the long-run fraction oftime that a system spends in a given subset of the energy surface is given by the measurethat Josiah Gibbs was to call microcanonical. See Uffink (2007: Sec.4.1 & Sec.6.1-6.2)for a good discussion of ergodic theory and Boltzmann’s ergodic hypothesis. For a classicdiscussion of Boltzmann’s hypothesis see Ehrenfest and Ehrenfest (2002).
69
be ergodic if and only if, for any set A, such that µ(A) > 0, the set of initial
points that never enters A has zero measure.30
An important part of the ergodic view is the following two-part result
established by George David Birkhoff (1931a,b).31
For any measure-preserving dynamical system,
1. For any A ∈ S, and where
χA(x) =
1, if x ∈ A
0, if x /∈ A,(3.12)
the limit
〈A, x0〉time = limT→∞
1
T
∫ T
0
χA(Tt(x0))dt (3.13)
exists for almost all points x0. That is, except for a set of measure zero.
2. If the dynamical system is ergodic, then
〈A, x0〉time = µ(A) (3.14)
for all A ∈ S and almost all x0 ∈ Γ.
The first part of the theorem says that it makes sense to talk about the
long-run fraction of time the system spends in A. Or rather, it makes sense
to talk about it for almost all initial microstates. The second part says that
the long-run fraction of time the system spends in A is equal to the phase
space volume of A—where we ignore those sets whose limit does not exist.
Again, noting the dominance of the equilibrium macrostate, it follows
30As usual, S is a σ-algebra on Γ.31It should be noted that this result is appropriated by supporters of typicality. There
is no indication that Birkhoff was or would be a supporter of typicality.
70
that almost all initial conditions lie on solutions that spend most of the time
in equilibrium and only show relatively short fluctuations away from it.
If we add to this result the assumption (or can show that) relevant sys-
tems are ergodic, and if we again interpret the measure on ΓE as a typicality
measure, and hold that it’s a good measure of what’s typical, then it follows,
given ordinary preparation procedures, that typical initial states approach
equilibrium, and that they remain in equilibrium for incredibly long peri-
ods of time. Again, just like the views that have come before, this view
could be seen as a way of fulfilling goals such as G1-4 without introducing
a temporally asymmetric element.
Standardly, this view is thought to be troubled by two problems.32 First,
it turns out to be extremely difficult to prove that anything that closely re-
sembles a realistic system is ergodic. As Frigg (2008b: p.124) notes, not even
a system of n elastic hard balls moving in a cubic box with hard reflecting
walls has been proven to be ergodic for arbitrary n. In fact, it has only
been proven in cases where n ≤ 4. What’s more, there are dynamical sys-
tems that exhibit thermodynamic behaviour that are provably not ergodic.
Jean Bricmont (2001), for example, highlights that the Kac ring model and
a system of n uncoupled anharmonic oscillators of identical mass exhibit
thermodynamic behaviour and that both are provably not ergodic. Other
well behaved, non-ergodic systems include: solids, and systems comprised of
non-interacting point particles.33 Examples such as these have led many to
32For discussions of these and other related problems, see Friedman (1976), Sklar(1995: Ch.5), Earman and Redei (1996), van Lith (2001), Emch and Liu (2002: Ch.7-9), Uffink (2007: Sec.6.1), and Frigg (2008b: Sec.3.2.4.3).
33See Uffink (1996: p.381) and Uffink (2007: p.1017). In a solid, say an ice cube, themolecules are tightly locked to their lattice site, and the phase point can access only a
71
think that ergodicity does not provide a satisfactory explanation of why iso-
lated systems spontaneously approach equilibrium and why they stay there
for long periods of time.34 The second worry standardly thought to trouble
the ergodic view is known as the “measure zero” problem. While, properly
speaking, the problem is usually levelled at a view closely related to what
we’re calling the ergodic view, in which the measure on ΓE is simply regarded
as a probability measure—whose probabilities require interpretation—rather
than as a typicality measure, it, in this context, effectively amounts to a
request to justify why we can neglect atypical microstates. Since Birkhoff’s
results apply to all but a set of states with measure zero, one would like to
have reason to neglect them. That is, one would like to have reason to treat
them as atypical states. The mere fact that the measure labels them as such
does not provide the justification that’s needed; sets that have measure zero
according to some measures have finite measure according to others. So, the
concern goes, what singles out this measure (or an equivalence class of mea-
sures) as the one to use to determine which sets are negligible? Since this is
really just an instance of a general worry that will be discussed later, further
discussion of it will be left till then.
minute region of the energy hypersurface.34See, for example, Earman and Redei (1996) and van Lith (2001). In fact, some—
often those who endorse something close to the unspecified dynamical view—take an evenstronger position on the ergodic enterprise. David Albert (2000: p.70), for example, thinksthat
. . . the prodigious effort that has over the years been poured into rigorousproofs of ergodicity is nothing more nor less—from the standpoint of thefoundations of statistical mechanics—than a waste of time.
72
3.2.5 The Epsilon Ergodic View
There is a view, or rather, an interpretation of a view, endorsed by some of
the discussions leading contributors, that’s closely related to the ergodic view.
It will be called the epsilon ergodic view. See Frigg and Werndl (2011, 2012a,
& 2012b).35 Unlike earlier views, this view restricts itself to accounting for
the behaviour of realistic gases. Its advocates do not claim to be offering
an account that speaks to the behaviour of macroscopic systems, generally.
The view is primarily aimed at overcoming the first of the worries standardly
levelled at the ergodic view. Its supporters hope to show that an impor-
tant class of realistic systems are epsilon ergodic, and that this dynamical
property, in conjunction with many of the ideas that have been introduced
previously, accounts for why they spontaneously approach equilibrium, and
why they spend large amounts of time in equilibrium.
To introduce epsilon ergodicity, it’s helpful to begin, as Frigg and Werndl
(2012b: p.104) do, by first defining ε-ergodicity—a distinct but related no-
tion. A dynamical system 〈ΓE, µE, φt〉 is ε-ergodic, where ε ∈ R and 0 ≤ ε <
1, if and only if there is a set Z ⊂ ΓE, with µ(Z) = ε, and with φt(ΓE) ⊆ ΓE
for all t ∈ R, where ΓE =df ΓE \ Z, such that the system 〈ΓE, µΓE, φΓE
t 〉 is
ergodic, where µΓE(·) =df µE(·)/µE(ΓE) for any measurable set in ΓE and
where φΓEt is φt restricted to ΓE. Naturally, a 0-ergodic system is simply an
ergodic system. We say that a dynamical system is epsilon ergodic if and
only if there exists a very small ε (i.e. ε << 1) for which the system is
35While Frigg and Werndl explicitly distance themselves from typicality in their 2011paper and again in their 2012b chapter, and regard themselves as offering a competingaccount, their work can be straightforwardly read through a typicality lens. What’s more,given their later work (see Frigg and Werndl 2012a, and Werndl 2013), it’s reasonable tothink that they would permit such a reading.
73
ε-ergodic.36
Since an epsilon ergodic system is ergodic on ΓE \ Z, the set of initial
states that lie in ΓE \Z spend most of the time in equilibrium and only show
relatively short fluctuations away from it. If ε is very small in comparison
to µE(ΓMP), then almost all initial microstates spend most of the time in
equilibrium and only show relatively short fluctuations away from it. Here,
“almost all” has been weakened to include all initial states except ones that
form a set of measure ε. If this measure is understood as a typicality measure,
then it follows that typical initial states exhibit thermodynamic behaviour.
Once again, this has been achieved without introducing a temporally asym-
metric element. So the result can be seen as a way of satisfying goals such
as G1-4.
Supporters of the epsilon ergodic view claim that the kinds of systems
usually called upon to show that ergodicity does not provide a sufficient
explanation for why isolated macroscopic systems spontaneously approach
equilibrium and why they remain in equilibrium for incredibly long periods
of time cuts no ice in this context. As Frigg and Werndl (2012b: p.107) argue,
Clearly, solids are not gases and hence can be set aside. Similarly,
uncoupled harmonic oscillators and the Kac ring model are irrel-
evant because they seem to have nothing to do with gases. The
properties of ideal gases are very different from the properties of
real gases because there are no collisions in ideal gases and colli-
sions are essential to the behaviour of gases. So while ideal gases
36Peter Vranas (1998) first introduced the concept of epsilon ergodicity to the founda-tions of statistical mechanics, but he put it to a different use.
74
may be expedient in certain context, no conclusion about the
dynamics of real gases should be drawn from them. Hence, the
well-rehearsed examples do not establish that there is a gas-like
system which behaves TD-like while failing to be ergodic.37
What’s more, they often cite a small collection of rigorous results along
with some numerical studies to support the idea that real gases are, in fact,
epsilon ergodic.38
3.3 Some General Concerns About
Typicality Accounts
This section discusses three worries commonly thought to trouble the typi-
cality accounts presented above. It also discusses, where possible, their most
promising solutions. Two of these worries were flagged earlier. The first
worry concerns interpreting Lebesgue measure as a typicality measure. The
second worry grants that it’s a typicality measure and asks, why should we
think it’s a good measure of what’s typical? The third worry concerns ex-
tending the results gained from the study of Boltzmann-like gases to other
37This conclusion is a bit odd. The conclusion Frigg and Werndl have in fact arguedfor is more narrow: that the well-rehearsed examples do not establish that there is a realgas system which behaves TD-like while failing to be ergodic. Their conclusion seems torely on a pretty dim view of idealised models and toy models. The Kac ring, for example,is thought to share salient features with real gases. In fact, it’s often called on, as it wasin Section 3.2.1, to highlight certain things about real gases because it has these features.And yet, Frigg and Werndl’s remarks imply that it’s not gas-like!
38See, for example, Frigg and Werndl (2011: Sec.7), Frigg and Werndl (2012b: Sec.7.6),and Frigg and Werndl (2012a: Sec.6).
75
macroscopic systems.39
3.3.1 Typicality Measures
All of the typicality accounts presented above trade on the assumption that
Lebesgue measure can be interpreted as a typicality measure. One might
wonder: what justifies interpreting this measure as a typicality measure?
A promising answer to this question has recently been offered by Charlotte
Werndl (2013).40 Werndl offers a set of conditions that measures ought to
satisfy so as to count as typicality measures and argues that these conditions
are satisfied by Lebesgue measure. This set builds on conditions that are
often implicitly, but sometimes explicitly, endorsed in the literature.
As mentioned earlier, a typicality measure represents the relative size of
sets of states. Any function T that describes the size of sets of states should
satisfy the standard axioms of a measure. That is, T (∅) = 0, T (A) ≥ 0 for
any measurable set A, and T (⋃i∈NAi) =
∑i∈N T (Ai) whenever Ai ∩ Aj = ∅
for all i, j, i 6= j. Since typicality measures are thought to represent the
relative size of sets of states, they are usually normalised, i.e., T (ΓE) = 1.
It was also mentioned earlier that typical states show a certain property if
the measure of the set that corresponds to this property is one or close to
one. Formally, typical states of ΓE have property P relative to a typicality
measure T if and only if T (ΓE \ D) ≤ β, where D ⊆ ΓE consists of all of
the states in ΓE that have property P and β is a very small real number,
39For a different discussion of these concerns, and other general concerns, see Frigg(2011: Sec.4.4). See Lazarovici and Reichert (2014: Sec.5) for a reply to the concernsraised by Frigg, and others.
40Pitowsky (2012) also attempts to justify typicality measures. See Werndl (2013: Sec.6)for a critique of this attempt.
76
possibly 0. D is called the typical set and ΓE \D is called the atypical set.
We’ll look at the conditions Werndl suggests for the case in which a typical
state possesses some property that it shares with a collection of states that
together form a set of measure one.41
The first condition Werndl (2013: p.474) notes is that the measure has
to be invariant under the dynamics. What’s typical at some time has to be
typical at both earlier and later times. Call this condition C1. This condition
is standardly endorsed by supporters of typicality.
C1. Typicality measures are invariant under the dynamics.
Since we’re unable to prepare systems into particular microstates, it’s
common to place a probability distribution (or class of distributions) over
the set of microstates compatible with our preparation procedures. Such a
distribution gives the probability that a system has been prepared in some
microstate. Werndl (2013: p.474) assumes that these probability distribu-
tions, p, are translation-continuous. That is,
lim‖τ‖→0
p(Tr(A, τ)) = p(A) for all open sets A, (3.15)
where ‖τ‖ is the standard Euclidean norm in Rn and Tr(A, τ) is the set
A translated by τ . That is,
Tr(A, τ) = Γ ∩ x+ τ |x ∈ A. (3.16)
41See Werndl (2013: Sec.8) for a set of conditions that are relevant for a more liberalkind of typicality measure; where typical states possess some property that they sharewith a collection of states that together form a set of measure close to one.
77
We’ll return to this assumption later. Next, Werndl (2013: p.475) assumes
that there is a class P of initial probability distributions of interest, which
she characterises as,
Distributions of Interest: The initial probability distributions
of interest are a class P of translation-continuous probability dis-
tributions where for every open set A there is a p ∈ P which
p(A) > 0.
The second part of this characterisation intends to capture the possibility
that, for any arbitrary open region of phase space, we cannot exclude the
possibility that there might be some way of preparing the system such that
there is a positive probability that it ends up in this region. While this
assumption does not impose a condition that a measure has to satisfy in
order to count, by Werndl’s lights, as a typicality measure, it does place a
restriction on the kinds of distributions with which these conditions have to
be satisfied.
The next two conditions are motivated by the idea that typicality mea-
sures should be related to initial probability distributions of interest. They
are:
C2. If p(A) = 0 for all probability distributions of interest for
some measurable A, then T (A)=0.
And,
C3. If T (A)=1 for some measurable A, then p(A) = 1 for all
probability distributions of interest.
78
C2 is the requirement that if a set of states has probability zero for all
initial probability distributions of interest, the set is atypical. C3 is the
requirement that whenever a set of states is typical, it has probability one
for all initial probability distributions of interest.
Werndl (2013: p.475) claims that a measure is a typicality measure when-
ever it satisfies conditions C1-3, for probability distributions characterised
as ones of interest. Werndl then argues that Lebesgue measure satisfies these
conditions, and so concludes that it can be interpreted as a typicality mea-
sure.42
While it may appear that the above considerations only provide us with
the resources to argue for the permissibility of interpreting Lebesgue measure
as a typicality measure, they, in fact, also provide us with the resources
to argue that it provides a good account of what’s typical. This is good
news because one might have worried first, that these measures were not
appropriately connected to experience, and so do not capture what’s typical
of actual systems, and second, that there are measures that can be placed on
ΓE that conflict with what’s typical according to Lebesgue measure.
The first of these worries is remedied by the connection of measures to
probability distributions of interest. They are related to our means of prepa-
ration of systems, and the rationale for assuming that they are translation-
continuous has to do with feasible preparation procedures—as we’ll see in a
moment. The second concern is avoided because of a result due to Malament
and Zabell (1980). Malament and Zabell have shown that a probability mea-
sure p is translation-continuous if and only if p is absolutely continuous with
42See Werndl (2013: Appendix A) for the details of this argument.
79
respect to the Lebesgue measure (p µ), where µ2 is said to be absolutely
continuous with respect to µ1 if and only if
if µ1(A) = 0 for a measurable set A, then µ2(A) = 0 (3.17)
for any two measures µ1 and µ2 defined on 〈Γ,S〉.
So then, the success of Werndl’s proposal effectively boils down to the
plausibility of her characterisation of probability distributions of interest,
and, relatedly, the assumption that initial probability distributions of interest
are translation-continuous. Regarding the later condition, Werndl cites the
motivation Malament and Zabell (1980) offer in support of this assumption.
They write:
Given two measurable sets on the constant energy surface, if one
is but a small displacement of the other, then it seems plausible
to believe that the probability of finding the exact microstate of
the system in one set should be close to that of finding it in the
other. (Malament and Zabell 1980: p. 346)
Unfortunately, this rationale does not uniquely support translation-continuity.
In the definition of translation-continuity, the perturbations of the set A are
ones such that every point of A undergoes the same displacement. But this
does not seem essential to the rationale offered by Malament and Zabell, and
it seems that the rationale should hold equally well for other transformations
of our measure space, whether or not each point gets the same displacement,
as long as the displacements are small. With this in mind, consider the
stronger condition of displacement-continuity, which we’ll define as follows.
80
A measure υ is displacement-continuous if and only if, for every
measurable set A, for any ε > 0, there exists δ > 0 such that
|υ(H(A))− υ(A)| < ε for all homeomorphisms H of the measure
space such that ‖H(x)− x‖ < 0 for all x.
This seems to be an intuitively reasonable condition to place on our prob-
ability measures. Unfortunately, it’s not satisfied by Lebesgue measure, and,
indeed, not satisfiable!43
While Malamant and Zabell’s remarks do not quite have the consequence
that’s needed, if one could motivate, on the basis of physical considera-
tions, translation-continuity, or some other continuity condition that’s equiv-
alent to being absolutely continuous with respect to Lebesgue measure, then
something that resembles the reasoning Werndl employs seems likely to go
through.
3.3.2 Extending Typicality Results
Typicality accounts attempt to extend results that apply to Boltzmann-like
gases to other macroscopic systems. Their success depends on the plausibility
of their extension. Notably, these accounts assume that relevant systems have
energy hypersurfaces that are dominated, like a Boltzmann-like gas, by their
respective equilibrium macrostate. They also assume that the microstates
that constitute the equilibrium macrostate are, like a Boltzmann-like gas,
typical. Both of these assumptions are dubious.
First, as Frigg (2011: p.89) notes, despite it often being stated as if it were
43See Luczak and Myrvold (unpublished) for more on this point, and for a discussion oftranslation-continuity, its rationale, and other continuity conditions.
81
a general truism, the result that the equilibrium macrostate dominates a sys-
tem’s energy hypersurface has only been proven for an ideal gas. That is, for
a system of non-interacting particles. What’s more, as Uffink (2007: p.976)
explains, it’s central to the proof that we’re dealing with an ideal gas. Since
most systems are not ideal gases, not even approximately, one may worry
about using this assumption to account for their behaviour. Moreover, as
Frigg (2011: p.90) also notes, it seems likely that this assumption is false, at
least for some thermodynamically well-behaved systems.44
Second, even if we grant that systems have energy hypersurfaces that
are dominated by their respective equilibrium macrostate, it doesn’t follow
that the measure of this set is one or close to one. In fact, it’s possible
that there are systems with distinct nonequilibrium macrostates that share
the same Boltzmann entropy value whose union has a measure greater than
the equilibrium macrostate. David Lavis has shown that this possibility is
realised by the Baker’s gas (a gas whose particles move according to the
Baker’s transformation) and the Kac ring model.45
Of course, most supporters of typicality are under no illusions about the
results they actually possess or about the ones they don’t. They are also not
under any illusions about what they can logically infer about the behaviour
of other systems on the basis of the results they actually possess. These
claims are especially true of those who support the unspecified dynamical
view. Many supporters of typicality seem to think of themselves as offering
arguments that are of heuristic value. They see themselves as offering a
44See also Callender (2010).45See Lavis (2005: pp.255-258) for a more detailed description of the Baker’s gas and
the relevant result, and see Lavis (2008: Sec.2) for the Kac ring model result.
82
general framework that highlights the form rigorous results would likely take.
Because of this, supporters are usually untroubled by these concerns. They
are also unmoved by those who demand more rigorous results; they often see
them as unnecessary.
3.4 The Limitations of Typicality Accounts
Suppose for the moment that we can treat Lebesgue measure as the appro-
priate and unique measure of typicality. Suppose also that every isolated
thermodynamic system has an energy hypersurface that is dominated by its
equilibrium macrostate. Or even just that many of the systems we’re inter-
ested in have this feature. Suppose further that this state has measure one
(or is close to one) and that the set of microstates that evolve under the
dynamics of the system that spend incredibly long periods of time in the
equilibrium macrostate also has measure one (or is close to one). Suppose
that by ordinary preparation procedures systems are placed into one of these
microstates with high probability. That is, with probability one or with a
probability close to one. And suppose that every isolated thermodynamic
system can be shown either to be ergodic or epsilon ergodic. Or even just
that many of the systems we’re interested in can. An aim of this section is
to highlight that typicality accounts have severe limitations, even if we grant
all of these things.
Typicality accounts do not provide us with the resources to answer im-
portant questions such as: what will this system do in the next five minutes?
Ten minutes? Hour? Year? What states will it pass through on its way to
83
equilibrium? If it reaches equilibrium, how long will it stay there before it
moves out of equilibrium? How likely is it that we’ll see the system fluctuate
out of equilibrium in the next few minutes? If it does, how large a fluctuation
should we expect? Etc. They do not underpin facts about the rates in which
systems approach equilibrium, or about the kinds of states they pass through
on their way to equilibrium, or about fluctuation phenomena. Interestingly,
it’s typically these other quantitative facts, and not the qualitative ones the
literature has focused on, that we care about most. Moreover, none of these
views help us form expectations about any of these quantitative facts or help
us justify the expectations we may already have about the behaviour of sys-
tems that begin away from equilibrium, having formed them on the basis
of experience. They also do not underpin quantitative relations such as the
fluctuation-dissipation theorem (see Sec.4.1).
Each view fails to underpin these facts, and fails to answer these questions,
for the same reason: they do not incorporate enough dynamical information.
The dominance view is unable to underpin these facts and answer these
questions because it does not incorporate any dynamical information. We
have no way of accounting for what a system will do without some account
of its dynamics. The unspecified dynamical view is unable to underpin these
facts, and answer these questions, because it, as its name suggests, leaves
a system’s dynamics, for the most part, unspecified. And both the ergodic
and epsilon ergodic views fail because they are tied to dynamical properties
that only hold in the long term time limit. Consequently, neither of these
properties can be used to validly conclude anything about a system’s state
84
at a particular time, or about its behaviour over finite time intervals.46
Of course, none of this should come as much of a surprise once we re-
member what typicality accounts hope to achieve. These accounts attempt
to show, in a way that satisfies goals such as G1-4, that isolated macroscopic
systems that begin away from equilibrium spontaneously approach equilib-
rium and that they remain in equilibrium for incredibly long periods of time.
Importantly, none of these goals makes any reference to the rates in which
systems approach equilibrium, or about the states they pass through on their
way to equilibrium, or about fluctuation phenomena. And why would they?
Not only would this undermine the generality of these claims, which is surely
one of their virtues, but it would also betray the macroscopic theory they’re
intended to be compatible with: thermodynamics. That is, a theory that
does not speak to any of these things. In fact, a theory no less, that despite
it’s somewhat misleading name, does not contain any dynamics.
In light of all of these shortfalls, it seems natural to turn some of our atten-
tion towards understanding and accounting for the success of the techniques
physicists use to model the behaviour of systems that begin away from equi-
librium. These techniques lead to equations that track a large and important
set of facts that concern a system’s behaviour. These equations also answer
the kinds of questions we’re often most interested in. Understanding and ac-
counting for the success of these techniques and equations would provide us
with an underpinning of a large and important set of quantitative facts that
typicality accounts cannot. Accounting for their success would also provide
46In fact, these criticisms also apply to views which hold that relevant systems aremixing. Some of these criticisms, however, do not apply to views which hold that relevantsystems are K-systems or Bernoulli. Again, for more on the ergodic hierarchy, see Berkovitzet al. (2006) and Frigg et al. (2014).
85
us with an underpinning, albeit a mosaic one, of the general qualitative facts
that are presently at the centre of foundational discussions. Here’s another
way of making these points. While there are certain qualitative facts we want
to understand about the behaviour of systems that begin away from equi-
librium, we also want to understand an associated set of quantitative facts,
and an account of how systems approach equilibrium quantitatively will a
fortiori tell us that they approach equilibrium and remain in equilibrium for
incredibly long periods of time.
What’s more, as an added bonus, this alternative line of research is more
in line with actual scientific practise. It’s worth noting that when physicists
model the behaviour of systems that begin away from equilibrium, they are
typically not, at the same time, attempting to satisfy goals such as the ones
that drive typicality accounts. While they are sensitive to the tension be-
tween the existence of irreversible macroscopic behaviour and an underlying
reversible dynamics, and while they are concerned about getting their models
to agree with experience, they are not concerned about expressing themselves
in ways that would underpin very general facts that concern the behaviour
of a wide variety of macroscopic systems. They are happy for their models,
which incorporate more specific dynamical information than the typicality
accounts presented above, to simply describe and predict the behaviour of
particular systems, or to describe and predict the behaviour of particular
classes of systems.
86
3.5 Summary
This chapter outlined and examined a collection of typicality accounts. These
accounts aim at explaining why isolated macroscopic systems spontaneously
approach equilibrium and why they remain in equilibrium for incredibly long
periods of time. They attempt to underpin these qualitative facts and satisfy
goals such as G1-4, thereby reconciling the existence of irreversible macro-
scopic processes with underlying reversible dynamics.
This chapter began by discussing Boltzmann’s H-theorem—which can
be seen as the first serious attempt to account for the behaviour of sys-
tems that begin away from equilibrium—and highlighted that it gives rise
to one of the very issues that is at the centre of contemporary philosophical
discussions of statistical mechanics: reconciling the existence of irreversible
macroscopic processes with underlying reversible dynamics. It then outlined
and discussed several typicality accounts. As was noted, these accounts are
inspired by, and build on, one of Boltzmann’s attempts to resolve the issues
that were sparked by his original derivation of the H-theorem. This chap-
ter considered some of the most pressing problems facing typicality accounts
and, where possible, discussed their most promising solutions. It then ex-
plained that these accounts have severe limitations, even when we ignore the
usual concerns. This chapter highlighted that they do not underpin a large
and important set of facts that concern the behaviour of systems that begin
away from equilibrium. This chapter suggested that this was a consequence
of what they aim at: underpinning two very general qualitative facts, while
satisfying goals such as G1-4. While typicality accounts may underpin some
important aspects of thermodynamics, they do not underpin many of the im-
87
portant quantitative facts that concern the behaviour of systems that begin
away from equilibrium. They are also of no help when it comes to forming
expectations about the rates in which systems equilibrate, or about the kinds
of states they pass through on their way to equilibrium, or about fluctuation
phenomena.
To remedy these shortfalls, this chapter ended by suggesting that we
pursue an alternative line of research. That we attempt to understand and
ground the success of the techniques and equations physicists commonly use
to model the behaviour of systems that begin away from equilibrium. This
line of research, if successful, has the potential to underpin not just some
important qualitative facts but also a large set of important quantitative
facts.
88
Chapter 4
Another Way to Approach the
Approach to Equilibrium
The previous chapter ended by highlighting the limitations of typicality ac-
counts. To remedy this shortfall, it promoted pursuing a different line of
research. That we attempt to understand and account for the success of the
techniques and equations physicists use to model the behaviour of systems
that begin away from equilibrium. The previous chapter indicated that by
understanding and grounding their success, we will not only be able to un-
derpin the qualitative facts the literature has focused on but we will also be
able to underpin many of the important quantitative facts that typicality
accounts cannot.
This chapter takes steps in this promising direction. It outlines and ex-
amines a technique commonly used to model the behaviour of an important
kind of system. More accurately, this chapter outlines and examines a tech-
nique commonly used to model the approach to equilibrium, and subsequent
89
behaviour, of a system in which a Brownian particle is introduced to an
isolated homogeneous fluid at equilibrium. As this chapter highlights, the
technique generates a collection of quantitatively accurate equations that
track important aspects of the system’s behaviour. This chapter also at-
tempts to account for the success of the model, by identifying and grounding
the technique’s key assumptions.
This chapter is composed of four sections. The first section outlines a way
of modelling the Brownian particle’s approach to equilibrium and subsequent
behaviour. As one might have expected, the approach takes its cue from the
theory of Brownian motion. It appeals to Langevin dynamics. The aim of
this section is to present the technique in a manner that reflects the way
physicists standardly model this kind of system.1 As this section highlights,
the approach generates a collection of interesting equations that track aspects
of the particle’s behaviour. These equations provide us with the resources to
answer the kinds of questions commonly asked about this kind of system. The
second section narrows in on, and discusses, the approach’s key assumptions.
The third section attempts to motivate a particular microphysical claim that
would ground the success of the technique, justify its key assumptions, and
underpin the facts the equations it leads to track. The fourth, concluding
1It should be noted that the system and technique discussed in this chapter draw ona range of assumptions supported by equilibrium statistical mechanics. While there arebenefits to considering systems and techniques that do not appeal to such assumptions,such as those that are described by, and that generate, Boltzmann’s equation, there are alsocosts. For one thing, these free-of-equilibrium-statistical-mechanical-assumption systemsand techniques are conceptually more difficult than the ones discussed in this chapter.Another reason to focus on the systems and techniques of Langevin dynamics are theirnovelty. There is already a wealth of philosophical literature that discusses Boltzmann’sequation. As far as the author can tell, no one, at least at the time of publication, hasdiscussed Langevin dynamics in the philosophical literature on statistical mechanics.
90
section, includes a summary and a suggestion about future research.
4.1 Modelling a Brownian Particle’s
Behaviour
The theory of Brownian motion is quite possibly the simplest approximate
way to treat the dynamics of nonequilibrium systems. The theory arose
from investigations into the irregular behaviour of objects such as pollen
grains and dust particles when they are placed into various kinds of fluids.
The phenomenon became widely known through the work of Robert Brown,
a botanist, who, in 1827, reported on the strange behaviour of pollen grains
floating on water. Albert Einstein (1905) and Marian Smoluchowski (1906)
provided the first theoretical analyses of Brownian motion. Since then, a
number of authors have helped develop the theory. Paul Langevin, a promi-
nent French physicist during the early parts of the 20th century, was one such
contributor.2 What’s particularly interesting about the theory of Brownian
motion is that it can be applied successfully to many other phenomena (e.g.
the motions of ions in water and the reorientation of dipolar molecules). In
fact, the theory has been extended to situations in which the “Brownian par-
ticle” is not really a particle at all, but is instead some collective property of
a macroscopic system. An example is the instantaneous concentration of any
component of a chemically reacting system near thermal equilibrium. In this
situation, the irregular fluctuations of the concentration in time correspond
to the irregular motion of the Brownian particle. What’s more, Brownian
2See, for example, Langevin (1908).
91
motion continues to be a topic of research nearly 200 years after Brown’s
work. As Pathria and Beale (2011: p.599) explain, much of the current in-
terest is due to the growth in the technological importance of colloids across
a wide range of fields, and the development of digital video and computer
image analysis.
The Langevin equation is the theory’s fundamental dynamical equation.
It contains both frictional forces and random forces. It’s a linear, first-order,
inhomogeneous differential equation. The Langevin equation is often used
to construct expressions that track interesting macroscopic variables. We
often have detailed quantitative information about these variables and they
capture what we’re often most interested in tracking. This section examines
the evolution of a system in which a Brownian particle is introduced to an
isolated homogeneous fluid at equilibrium. It also highlights the Langevin
equation’s use in the construction of equations that track the behaviour of
two interesting macroscopic variables—the Brownian particle’s mean squared
displacement and its mean squared velocity. The theory also possesses a
fluctuation-dissipation theorem. It relates the forces that appear in the
Langevin equation to each other. This theorem has many important and
far-reaching generalisations. This section includes an elementary version of
the theorem.
When a single Brownian particle is immersed in an isolated homoge-
neous fluid at equilibrium it typically does one of three things, depending
on its initial velocity. If the Brownian particle enters the fluid with a mean
squared velocity (an expectation value around which the actual value fluctu-
ates) that’s greater than the value given by the equipartition theorem, then
92
it typically slows down to that value. If the Brownian particle enters the fluid
with a mean squared velocity that’s less than the value given by the equipar-
tition theorem, then it typically speeds up. The particle’s speeding up and
slowing down is the result of collisions with the molecules that comprise the
fluid. If the Brownian particle enters the fluid with a mean squared velocity
that’s close to the value given by the equipartition theorem, then it typically
remains at this value. When the Brownian particle reaches equilibrium, it
jostles about, on macroscopic time scales, with a mean displacement of zero.
Slight deviations from this behaviour can occur, but large deviations are ex-
tremely rare. From the standpoint of an ordinary observer, the Brownian
particle’s short-term behaviour seems to be random, both as it approaches
equilibrium and once it has reached equilibrium. As such, there isn’t much
more an unaided observer can say about its motion.
While the motion of a Brownian particle appears to be random, it’s
nonetheless describable by the same equations of motion that model the
behaviour of other classical systems.3
We’ll consider the one-dimensional motion of a spherical particle with
radius a, mass m, position x, and velocity V , in a fluid medium with viscosity
η.
Newton’s equation of motion for the particle is
3One might argue that it’s inappropriate to be talking about classical mechanics inconnection with the Langevin equation, since the wave packets of molecules will spreadtoo quickly for a classical approximation to be appropriate.
However, in an appropriate regime, even if we can’t treat the molecules as approximatelylocalised, the Wigner function will, to a good approximation, obey the same Liouville equa-tion that the classical probability density function obeys, and hence thinking about whatclassical trajectories will do is a useful way of visualising the behaviour of the quantumprobability density. The expectation values we consider can be taken to be expectationvalues of quantum observables, and much the same things will be said in that context.
93
Figure 4.1: The mean squared velocity of a Brownian particle as a functionof time. Curves 1,2, and 3 correspond, respectively, to the initial conditionsv2 = 6KT/m, 3KT/m, and 0. τ is the characteristic time. (Pathria and
Beale 2011: p.591)
mdv
dt= Ftotal(t). (4.1)
Ftotal(t) is the total instantaneous force on the particle at time t. This
force is the result of the particle’s interaction with the fluid molecules. If the
positions of the molecules are known as functions of time, then, in principle,
this force is a known function of time. In this sense, the particle is not subject
to a “random force” at all.
It’s usually not practical or even desirable to look for an exact expression
of Ftotal(t). Experience teaches us that in ordinary cases this force is domi-
nated by a frictional force, −ζV , that is proportional to the velocity of the
Brownian particle. The friction coefficient is given by Stokes’ law, ζ = 6πηa.
94
If this were the whole story, the equation of motion for the Brownian particle
would be
mdV
dt∼= −ζV, (4.2)
whose solution is
V (t) = e−ζt/mV (0). (4.3)
According to (4.3), the velocity of the Brownian particle decays, over
time, to zero. This result, however, cannot be quite right. It follows from
the equipartition theorem that the mean squared velocity of the particle at
thermal equilibrium is 〈V 2〉eq = KT/m. Obviously, the assumption that
Ftotal(t) is dominated by the frictional force has to be modified.
Standardly, an additional force, δF (t), is added to the frictional force.
For reasons that will become clear shortly, this force is referred to as the
fluctuating force. The equation of motion then becomes
mdV
dt= −ζV + δF (t). (4.4)
This is the Langevin equation for a Brownian particle. It partitions the
total force in two. Because both parts are the result of the Brownian particle’s
interaction with the fluid, it’s reasonable to think that a fundamental relation
exists between them.
The additional force, δF (t), is the result of continual collisions between
the Brownian particle and molecules in the fluid. It’s typically assumed,
given the observed randomness of an individual trajectory, that this force
95
has the following properties.
〈δF (t)〉 = 0, (4.5)
and
〈δF (t)δF (t′)〉 = 2Bδ(t− t′). (4.6)
Because of these properties, δF (t) is standardly referred to as the fluctu-
ating force. B is a measure of the strength of the fluctuating force. The delta
function in time indicates that there is no correlation between collisions in
distinct time intervals dt and dt′.
The Langevin equation can be solved to give
V (t) = e−ζt/mV (0) +
∫ t
0
dt′e−ζ(t−t′)/mδF (t′)/m. (4.7)
The first term describes the exponential decay of the particle’s initial
velocity. The second term describes the extra velocity produced by the fluc-
tuating force.
This expression can be used, in conjunction with (4.5) and (4.6), to model
the equilibration of the Brownian particle’s mean squared velocity. There are
three contributions to V (t)2. The first one is
e−2ζt/mV (0)2. (4.8)
It decays to zero at long times. Next are two cross terms, each first order
in the noise,
96
2V (0)e−ζt/m∫ t
0
dt′e−ζ(t−t′)/mδF (t′)/m. (4.9)
When (4.5) is used to average over the noise, these terms vanish. The
final term is second order in the noise:
∫ t
0
dt′e−ζ(t−t′)/mδF (t′)
∫ t
0
dt′′e−ζ(t−t′′)/mδF (t′′)/m2. (4.10)
Now the product of the two noise factors is averaged, according to (4.6).
This leads to
∫ t
0
dt′e−ζ(t−t′)/m
∫ t
0
dt′′e−ζ(t−t′′)/m2Bδ(t′ − t′′)/m2. (4.11)
The delta function removes one time integration, and the other can be
done directly. The resulting mean squared velocity is
〈V (t)2〉 = e−2ζt/mV (0)2 +B
ζm(1− e−2ζt/m). (4.12)
In the long time limit, the exponentials fall out, and this quantity ap-
proaches B/ζm. At long times the mean squared velocity obtains the value
KT/m, as given by the equipartition theorem. As a consequence, we find
that
B = ζkT. (4.13)
This result is known as the fluctuation-dissipation theorem. It relates the
strength B of the fluctuating force to the magnitude ζ of the friction.
The evolution of one of the system’s macrovariables is described by (4.12).
97
It models the system’s spontaneous approach to equilibrium. It provides
qualitative and quantitative information. It tells us to expect the system to
approach equilibrium reasonably quickly and for it to remain in equilibrium
for incredibly long periods of time. It helps us form expectations about the
kinds of macrostates the system will be in at various stages of its evolution
and it helps us form expectations about how quickly the system will reach
equilibrium. It also helps us form expectations about fluctuation phenomena.
What’s more, as the fluctuation-dissipation theorem reveals, (4.12) also leads
to quantitative information about the strength of the fluctuating force.
The Langevin equation can also be used, again with the help of (4.5)
and (4.6), to derive an expression for the mean squared displacement of a
Brownian particle that’s been in the fluid for a long time.
Multiplying both sides of the Langevin equation by x gives
mxdV
dt= m
[ ddt
(xV )− V 2]
= −ζxV + xδF (t). (4.14)
If we take the average of both sides and use (4.5), we find that
m〈 ddt
(xV )〉 −m〈V 2〉 = −ζ〈xV 〉. (4.15)
From the equipartition theorem we have 12m〈V 2〉 = 1
2KT , so (4.15) be-
comes
m〈 ddt
(xV )〉 = KT − ζ〈xV 〉. (4.16)
This is a first-order differential equation that can be solved using the
integrating factor technique. Its solution is
98
〈xV 〉 = Ce−ζmt +
KT
ζ, (4.17)
where C is a constant of integration. Let τ ≡ ζm
, so that τ−1 is a charac-
teristic time constant of the system. If we take as our initial condition that
x = 0 at t = 0, then (4.17) entails that C = −KTζ
. So we have
〈xV 〉 =KT
ζ(1− e−τt). (4.18)
If we replace 〈xV 〉 by 12ddt〈x2〉 and integrate (4.18) with respect to t, we
obtain
〈x2〉 =2KT
ζ[t− 1
τ(1− e−τt)]. (4.19)
(4.19) gives the mean squared displacement of the Brownian particle.
Note two interesting limiting cases. Case 1 : t << τ−1. If t < τ−1, then
using a Taylor series expansion,
e−τt = 1− τt+1
2τ 2t2 − . . . . (4.20)
So for t << τ−1 and O(t3) = 0,
〈x2〉 =KT
mt2. (4.21)
This means that on very short time scales the particle behaves as if it
were a free particle moving along with constant velocity V = (KT/m)12 .
99
Case 2: t >> τ−1. If t >> τ−1, then e−τt → 0. So (4.19) simply becomes
〈x2〉 =2KT
ζt. (4.22)
(4.19) provides both interesting qualitative information and interesting
quantitative information about one of the system’s macrovariables. It models
the Brownian particle’s behaviour. The Brownian particle jostles about with
a mean displacement of zero and with a mean squared displacement given by
(4.19). It tells us to expect it to behave as if it were a free particle moving
along with constant velocity on very short time scales, and for it to behave
like a diffusing particle executing a random walk on longer time scales. It
helps us form expectations about the kinds of macrostates the system will
be in at various stages of its evolution, and it helps us form expectations
about the Brownian particle’s fluctuations. And this does not exhaust the
quantitative information that can be extracted from the model. Since we’re
able to measure 〈x2〉 experimentally, then, if we know the size and density of
the particles, as well as the viscosity of the medium, we can deduce from these
observations a good approximation of K, Boltzmann’s constant. Moreover,
if we have knowledge of the gas constant, then we can also calculate the value
of Avogadro’s number.
As a final point, consider the results of the following real study. In the
study, experimenters observed the behaviour of a spherical particle immersed
in water and recorded aspects of its Brownian motion. The study highlights
the predictive accuracy of (4.19) and (4.22). For the complete details of
the study, see Lee, Sears, and Turcotte (1973). In the study, T ' 300K,
η ' 10−2 poise, and a ' 4× 10−5cm. The study found that 403 values of the
100
net displacement, ∆x, of the particle, observed after successive intervals of 2
seconds each, were distributed as follows:
Table 4.1: Observed displacements of a spherical particle immersed inwater. (Pathria and Beale 2011: p.591)
The data indicates that 〈x2〉 = 2.09 × 10−8cm2. This observationally
calculated result is remarkably close to the value given by (4.22).
4.2 Some Comments on the Approach
The purpose of this section is to narrow in on the approach’s key assump-
tions and to set aside aspects of the approach that may appear mysterious.
Justifying these assumptions is central to underpinning the qualitative and
quantitative facts the technique tracks.
101
4.2.1 Constructing a Langevin Equation
First, it may seem as though the Langevin equation came out of nowhere. Its
form was motivated by aspects of ordinary experience. The discussion began
by noting that experience teaches us that in ordinary cases the total force
acting on the particle is dominated by a frictional force, −ζV , proportional to
the velocity of the Brownian particle. It was then indicated that this couldn’t
be the whole story, since it’s at odds with what we observe over longer periods
of time. This led to adding a force to the frictional force that would account
for the difference. We were thus led to (4.4), the Langevin equation for a
Brownian particle. This kind of motivation is common.4 It is, however, “top
down”. Since it’s motivated by phenomenal considerations, one may worry
about whether and how it connects to the microphysics. Happily, it can be
constructed from a combination of physical and mathematical considerations.
Consider a similar but simpler system than the one discussed in the pre-
vious section, a one-dimensional system that’s comprised of a relatively large
Brownian particle with mass M , which is hit from both sides by molecules
of mass m.5 We’ll also assume that M m and that collisions are elastic.
The velocity of the Brownian particle before and after a single collision will
be denoted by V and V ′, respectively, and the velocity of a molecule before
and after a collision will be denoted by v and v′, respectively. If we combine
the equations for conservation of momentum and energy, then we can write
the velocities after the collision in terms of the velocities before the collision.
That is,
4See Reif (1965: Sec.15.5), Zwanzig (2001: Ch.1), Mazenko (2006: Ch.1), and Pathriaand Beale (2011: Ch.15), for example.
5This system is also discussed by de Grooth (1999).
102
V ′ =M −mM +m
V +2m
M +mv (4.23)
and
v′ =m−MM +m
v +2M
M +mV. (4.24)
Using the assumption that M m, and the following approximations,
M −mM +m
≈ 1− 2m
M+O
((mM
)2), (4.25)
M
M +m≈ 1− m
M+O
((mM
)2), and (4.26)
m
M +m≈ m
M+O
((mM
)2), (4.27)
(4.23) can be written as
V ′ =
(1− 2m
M
)V +
2m
Mv. (4.28)
So the change in momentum of the Brownian particle due to a single
collision is
∆P = 2mv − 2mV. (4.29)
And the momentum change of the Brownian particle due to N collisions
is
103
∆PN = 2mN−1∑i=0
vi − 2mN−1∑i=0
Vi. (4.30)
If we consider a time interval ∆t that’s small enough that the velocity of
the Brownian particle does not change appreciably, but because M >> m,
we still have a large number of collisions, then the second sum in (4.30) can
be approximated by 2mNV = 2mnV (t)∆t, where V (t) is the velocity of
the Brownian particle at time t, and n is the mean number of collisions per
second so that N = n∆t. So, we have,
∆PN = 2mn∆t−1∑i=0
vi − 2mnV (t)∆t. (4.31)
Dividing both sides by ∆t yields an expression for the time derivative of
the Brownian particle’s velocity.
MdV
dt= −γV + Fs, (4.32)
where the damping constant γ = 2mn and
Fs =1
∆t
n∆t−1∑i=0
2mvi. (4.33)
Fs is a fluctuating force, if, like before, it’s assumed to have the following
properties:
〈Fs(t)〉 = 0, (4.34)
and
〈Fs(t)Fs(t′)〉 = 2Bδ(t− t′). (4.35)
104
(4.34) entails that the incoming velocities of colliding molecules have an
expectation value of zero. (4.35) entails that the incoming velocities of col-
liding molecules at distinct times are uncorrelated.
(4.32) has the same form as the Langevin equation. It, however, contains
explicit expressions for both the damping force and the fluctuating force.
Much like (4.4), (4.32) can be used together with (4.34), (4.35), and the
equipartition theorem, to arrive at important results such as (4.22).
4.2.2 The Time-Reversal Non-Invariance of the
Langevin Equation
Something else that may have seemed mysterious about the approach de-
scribed in the previous section is the time-reversal non-invariance of the
Langevin equation. The one-dimensional version, (4.32), is also asymmetric
under time-reversal. This too may seem mysterious. Each equation describes
the evolution of a system whose underlying dynamics are symmetric under
time-reversal.
To see that (4.32) is asymmetric under time-reversal, first note that the
expectation value of the fluctuating force is zero. Then, taking the expecta-
tion values of both sides of (4.32), we get the conclusion that the expectation
value of V decays exponentially to zero. The temporal reverse of this would
have an exponentially increasing expectation value of V . This reasoning also
applies to the Langevin equation. Again, the expectation value of the fluctu-
ating force is taken to be zero. Since the frictional force term is temporally
asymmetric, we are led, when we take the expectation values of both sides of
(4.4), to the conclusion that the expectation value of V decays exponentially
105
to zero. And similarly, the temporal reverse would have an exponentially
increasing expectation value of V .
The time-reversal non-invariance of the Langevin equation is the result
of the temporal asymmetry of its frictional force term and the properties
attributed to its second, fluctuating force term. The same is true of (4.32).
Interestingly, both equations, understood simply as equations that divide the
total force acting on the Brownian particle in two, are compatible with time-
reversal invariance. What’s more, this compatibility holds even after the
choice is made to split the total force into a frictional force and what is left
over. What ensures the time-reversal non-invariance of these equations, given
the time-reversal non-invariance of their respective frictional force terms, are
the properties attributed to their second terms.
The time-reversal non-invariance of the Langevin equation prompts fur-
ther questions. First, why do physicists standardly assume that the Langevin
equation’s second term has the properties identified in (4.5) and (4.6)? And
second, what does their rationale reveal about the microphysics of those sys-
tems whose behaviour is accurately modelled by the Langevin equation, and
the equations derived from it (e.g. (4.12) and (4.19))? Otherwise said, what
would have to be true, or at least be approximately true, at the microlevel to
account for the success of equations such as (4.4), (4.12), and (4.19), given
that the underlying dynamics is symmetric under time-reversal?
First, it’s standardly assumed that the second force term has an expecta-
tion value of zero because the Brownian particle, observed at ordinary scales,
seems just as likely to move in any direction at any moment as any other.
In the one-dimensional model, the analogous thought is that the Brownian
106
particle seems just as likely to move, at any moment, to the left as it is
to the right. What’s more, if observation provided us with reason to think
that the Brownian particle was more likely to move in one direction than
another—say, because we thought it was more likely to be struck from one
side rather than another—then we would describe this thought using some
force expression and extract it from the second term.
Interestingly, (4.5) is standardly presented alongside an assumption anal-
ogous to (4.6).6 Notice too that (4.34) was presented along with (4.35).
Both (4.6) and its analogue, (4.35), say that no correlation exists between
the forces acting on the Brownian particle at any two distinct times.7
Importantly, what standardly underlies and motivates both (4.5) and
(4.6), or, analogously, (4.34) and (4.35), is something that closely resembles
the following microphysical assumption.
Collision Assumption: That v is, at any time, independent of
V .
Spelled out, the underlying claim is that the velocity of any incoming
colliding fluid molecule (drawn from the distribution of velocities of molecules
of the fluid), is, at any time, probabilistically independent of the incoming
velocity of the Brownian particle.
Interestingly, the collision assumption is temporally asymmetric. To see
this, consider the one-dimensional model. Next, note that the collision as-
6See Zwanzig (2001: p.5), Kadanoff (2000: p.120), and Mazenko (2006: p.8), for exam-ple.
7It’s worth noting that concerns about the conflict between the time-reversal non-invariance of the Langevin equation and the time-reversal invariance of the underlyingdynamics do not standardly play a role in the determination of properties attributed tothe Langevin equation’s second term.
107
sumption has to do with the probability distribution of the velocities of
molecules that are about to hit the Brownian particle. Now consider (4.24).
In particular, consider the probability distribution of v′. That is, the proba-
bility distribution of the velocities of molecules that have just collided with
the Brownian particle. If the expectation value of v is zero, as (4.34) entails,
then the expectation value of v′ is proportional to V . So, if V is greater than
its equilibrium value, then, on average, colliding molecules that are in the
path of the Brownian particle’s motion will have a greater velocity after the
collision than before. Meanwhile, colliding molecules that are travelling on
the Brownian particle’s path will, on average, have a post-collision velocity
that’s less than their pre-collision velocity. When we velocity-reverse the
situation, we get a state of affairs in which the distribution of velocities of
incoming colliding molecules are, contrary to the assumption, not indepen-
dent of the Brownian particle’s velocity. They appear conspiratorial. The
ones travelling on its path will, on average, have a pre-collision velocity that’s
greater than the average molecule and the ones traveling in its path will, on
average, have a pre-collision velocity that’s less than average.
4.2.3 Why Does the Langevin Equation Work?
Since (4.4), (4.12), and, in particular, (4.19) have consequences that are con-
firmed by experiment, it’s worth considering what would have to be true, or
be at least approximately true, at the microlevel, to justify the use of the
collision assumption, and to ensure that its consequences—(4.5) and (4.6)—
are at least approximately true. Answering this helps explain the success
of equations such as (4.4), (4.12), and (4.19). It also helps to underpin the
108
quantitative and qualitative facts these equations track. As one might have
expected, the answer is suggested by the assumption itself. The natural sug-
gestion is that it’s a convenient approximation of the following microphysical
fact.
Microphysical Fact: That v is at most independent of V at
some initial time and that at any other time v is effectively inde-
pendent of V .
The claim here is that, at all times (except, perhaps, initially), the velocity
of any incoming colliding fluid molecule and the incoming velocity of the
Brownian particle are approximately probabilistically independent.8
Notice that we don’t claim that the velocities of incoming colliding molecules
are, at all times, in fact independent of the incoming velocity of the Brownian
particle, as the collision assumption has it. The reason for this more modest
claim is that these velocities are at least correlated for very short times after
collision. This can be seen by reflecting on what we think’s going on at the
microlevel. Once the Brownian particle is introduced to the fluid, it begins
to collide with fluid molecules. The subsequent positions and momenta of
the molecules, like the Brownian particle, are the result of collisions. Because
the forces acting on the particle can be expressed by smooth functions—at
least to a good approximation—the forces resulting from these collisions will
be correlated.
But it should be remembered that the motion of the Brownian particle
appears to be random on macroscopic time scales. So it seems reasonable to
8The idea then is that, for practical purposes, one can treat the velocities of these col-liding particles as if they are probabilistically independent. That is, one may be justifiedin using the collision assumption.
109
think, as the microphysical fact intends to suggest, that any correlations that
form between the velocities of colliding particles wash out incredibly quickly.
Happily, weakening (4.6) (or (4.35)) to accommodate these considerations—
say by instead assuming that correlations between fluctuating forces at dis-
tinct times exponentially decay on scales that are incredibly short compared
to the system’s relaxation to equilibrium—yields results that closely approx-
imate (4.19) and (4.13). Similar results also hold if we assume that correla-
tions between fluctuating forces at distinct times are Gaussian on scales that
are extremely short compared to the system’s relaxation to equilibrium.
Having identified the approach’s crucial assumption, the collision assump-
tion, it’s important to ask whether its use is justified. Or, more crucially,
what reason do we have to think that the microphysical fact is, in fact,
true? This question is important, since we can hardly ground the success of
the technique, or the facts its resulting equations track, on an unsupported
microphysical claim.
4.3 Motivating the Collision Assumption
The microphysical fact, if true, provides reason to utilise the collision as-
sumption. Support for the microphysical fact comes from at least two sources:
from a scientifically informed reflection on the microphysical behaviour of rel-
evant systems, and from results that emerge from the study of an idealised
system, a hard sphere gas.
First, when thinking about the behaviour of a Brownian particle that’s
been placed into an isolated homogeneous fluid at equilibrium, it seems rea-
110
sonable to have the following picture in mind. Once the Brownian particle is
placed into the fluid, it begins to collide with fluid molecules as it approaches
equilibrium. Since the fluid is composed of a very large number of molecules,
which are involved in very many collisions, it’s reasonable to think that when-
ever two of them collide it’s overwhelmingly likely that a very large number
of collisions will occur before they collide again—if, in fact, they ever collide
again. This suggests that by the time the Brownian particle were to col-
lide with some molecule it had already encountered neither would, in effect,
carry a memory of their past collision. That is, any correlation that resulted
from their previous interaction would have effectively been washed out. Be-
cause these collisions happen incredibly quickly, so too would the washing
out of correlations. So then, we can reasonably expect the forces acting on
the Brownian particle to be, at any moment, effectively random, and we
can expect the velocities of incoming colliding molecules to be, at any mo-
ment, effectively independent of the velocity of the Brownian particle. The
reasonableness of this picture supports the truth of the microphysical fact.
Notice too how bizarre the temporal reverse of a situation in which a
Brownian particle slowed to equilibrium would be. That is, a situation in
which the microphysical fact did not hold. If we reversed the velocities of all
of the molecules that comprise the system, then we would witness a conspir-
atorial situation similar to the one described at the end of the last section.
The Brownian particle would miraculously speed up and the velocities of the
incoming colliding molecules would be correlated with it. The molecules that
are about to hit the Brownian particle from behind would, on average, have
a higher than average velocity, and the ones that are in its path would, on
111
average, have a lower than average velocity.
What makes the first picture reasonable and the second situation bizarre
is that only in the second situation are the velocities of incoming colliding
molecules correlated before they interact. What’s peculiar about this is that
in the absence of some earlier event that would account for this correlation,
their correlation is to be explained by some event that lies in their future.
And this flies in the face of the fundamental idea that causes precede their
effects.9
If we assume that the microphysical fact holds (or that something close
to it holds), then we are led to equations that accurately model and predict
the behaviour of systems we can and do observe. The models outlined in
earlier sections attest to this fact. So we have support for the truth of the
microphysical fact. As the discussions of the temporal reversals of the ordi-
nary behaviour of these systems suggest, deviations from such an assumption
lead us to expect systems to behave in ways we simply do not ever witness.
So we have further support for the truth of the microphysical fact.
The study of idealised models also supports the truth of the microphysical
fact. Oscar Lanford III has shown that an instance of the microphysical fact
is true, under certain conditions, for a hard sphere gas model.10 Lanford has
shown that when the initial state of the gas satisfies a certain condition—that,
in effect, amounts to there being a lack of correlation between the momenta
9This claim might be controversial to some. Huw Price (1996), for example, disputesits truth at the microphysical level. The following things, however, are not controversial.First, we take this idea for granted in our ordinary dealings with the world. That it’suniversally true is the default position. Second, even on reflection, it’s not disputed at themacroscopic level. Third, scientists standardly develop successful physical models on thebasis that causes precede their effects.
10See Lanford III (1975, 1976, & 1981).
112
and positions of gas molecules—the system effectively sustains this property,
as it approaches equilibrium.11 The result, which follows from Lanford’s
theorem, applies to systems whose behaviour is described by Boltzmann’s
equation. These systems also have an underlying dynamics that is invariant
under time-reversal. While the result holds only for a short period of time,
there are good reasons to think that it can be extended.12
On the basis of these considerations, it seems reasonable to think that the
microphysical fact is true. Its truth would support the use of the collision
assumption and its consequences, (4.5) and (4.6) ((4.35) and (4.35)). In
particular, it would justify the use of (4.5) and (4.6) in the derivations of
(4.12) and (4.19). Its truth would also help explain the success of these
equations, and help underpin the quantitative and qualitative facts these
equations track.
4.4 A Summary and a Suggestion
The last chapter suggested pursuing a particular line of research. More ac-
curately, it promoted turning some of our attention towards understanding
and grounding the success of the techniques and equations physicists use
to model the behaviour of systems that begin away from equilibrium. This
chapter took steps in this direction by outlining, examining, and attempt-
ing to ground the success of a technique and equation used to model the
behaviour of a Brownian particle that’s been immersed in a homogeneous
11See Uffink (2007: Sec.6.4), Uffink and Valente (2015), and Valente (2014) for discus-sions of Lanford’s results.
12For a discussion of its extension, see Valente (2014: Sec.7.2) and Lanford III(1976: p.14).
113
fluid. This chapter highlighted that the technique generates a collection of
interesting and quantitatively accurate equations whose predictions are con-
firmed by experiment. This chapter attempted to account for the success of
the model, by identifying and motivating the technique’s key assumptions.
It noted that support for them can be traced back to an endorsement of the
collision assumption, which is a convenient approximation of the microphys-
ical fact. The body of the chapter ended with several reasons to think that
the microphysical fact is, in fact, true.
Of course, the technique and equation discussed in this chapter are but
the first in a collection of more elaborate and predictively accurate techniques
and equations.13 There are also limitations on the model’s applicability—
despite its generality. Naturally, it would be good for future investigations to
consider and examine more elaborate techniques and equations, and for them
to also examine more complicated systems (e.g. systems that are influenced
by external forces). It would be good for these investigations to uncover and
discuss the presuppositions these techniques and equations trade on (if any),
and, where possible, to discuss the grounds we have for believing that these
conditions are satisfied.
A more concrete suggestion is to examine and motivate the steps that
lead to the Fokker-Plank equation—a probabilistic version of the Langevin
equation. The Fokker-Plank equation is the starting point for many useful
calculations in nonequilibrium statistical mechanics. For example, it’s used
to determine the rate at which a Brownian particle crosses a potential bar-
rier (Zwanzig 2001: p.40). The Fokker-Plank equation is often called upon
13See, for example, Zwanzig (2001) and Mazenko (2006).
114
in situations in which it’s suitable to model a system’s behaviour using a
non-linear Langevin equation. Linear Langevin equations are easy to solve
analytically, non-linear Langevin equations are not. Physicists standardly
side-step this difficulty by constructing a Fokker-Plank equation that corre-
sponds to a given Langevin equation. It would be good to highlight whatever
presuppositions are required to make these steps legitimate, and to discuss
the grounds we have for believing that these conditions are met.
More obviously though, it would be good for future investigations to
return to Boltzmann’s equation (and the many versions of it) and to examine
and motivate the steps that lead to it. Boltzmann’s equation accurately and
quantitatively models the behaviour of dilute gases that begin away from
equilibrium. This fact is often lost amidst talk of its problems.14 As the
discussion of Boltzmann’s equation in Section 3.1 and the underpinning of the
Langevin equation offered in Section 4.3 suggest, this research project would
involve an attempt to motivate an assumption similar to the Stoßzahlansatz.
The successful grounding of these, and other techniques and equations,
would likely underpin many of the important and interesting quantitative
facts that typicality accounts cannot. Accounting for the success of these
techniques and equations would also lead, in a mosaic fashion, to an under-
pinning of the very concepts that are currently at the centre of foundational
discussions of statistical mechanics.
14See also Wallace (2013: p.14).
115
Chapter 5
Conclusion
This thesis commented on aspects of foundational debates that arise in con-
nection with statistical mechanics and thermodynamics.
One of its overarching goals was to promote a particular line of research:
understanding and accounting for the success of the techniques and equa-
tions physicists use to model the behaviour of systems that begin away from
equilibrium.
This thesis helped explain what philosophers mean when they say that
an aim of nonequilibrium statistical mechanics is to account for certain as-
pects of thermodynamics, by making the issue of reconciling the existence
of thermodynamically irreversible processes with underlying reversible dy-
namics more clear. This contributed to the overarching goal of promoting
a particular line of research by making clear the goals philosophers have
traditionally set themselves. This enabled us, in later chapters, to better ap-
preciate and evaluate their attempts to achieve them, and to better situate
and appreciate how the proposed line of research fits with the contemporary
116
literature.
As was noted, many of the leading attempts to reconcile the existence of
thermodynamically irreversible processes with underlying reversible dynam-
ics proceed by way of, and are typically part and parcel of, discussions that
attempt to underpin the following qualitative facts: (i) that isolated macro-
scopic systems that begin away from equilibrium spontaneously approach
equilibrium, and (ii) that they remain in equilibrium for incredibly long peri-
ods of time. As we saw, these attempts appeal to phase space considerations
and notions of typicality. This thesis examined leading typicality accounts
and highlighted their severe limitations. For one thing, they do not underpin
a multitude of interesting quantitative facts that arise in connection with
the behaviour of systems that begin away from equilibrium. To remedy this
and other shortfalls, this thesis promoted working on the alternative project,
since accounting for the success of the techniques and equations physicists
use to model the behaviour of systems that begin away from equilibrium
helps underpin not just the qualitative facts mentioned above, which the lit-
erature has focused on, but also many of the important quantitative facts
that typicality accounts cannot.
This thesis also took steps in this promising direction. The fourth chapter
outlined, examined, and attempted to ground the success of a technique com-
monly used to model the approach to equilibrium and subsequent behaviour
of a Brownian particle that’s been introduced to an isolated homogeneous
fluid at equilibrium. It attempted to account for the success of the model,
by identifying and grounding the technique’s key assumptions. The chapter
ended by offering some suggestions about where this research may be taken
117
next. It suggested examining more elaborate techniques and to ground their
success along with the predictively accurate equations they help generate.
Bibliography
Albert, David Z. 2000. Time and Chance. Harvard University Press.
Berkovitz, Joseph, Roman Frigg, and Fred Kronz. 2006. “The Ergodic Hi-
erarchy, Randomness and Hamiltonian Chaos”. Studies in History and
Philosophy of Science Part B: Studies in History and Philosophy of Mod-
ern Physics 37: pp. 661–691.
Birkhoff, G. D. 1931a. “Proof of a Recurrence Theorem for Strong Transitive
Systems”. Proceedings of the National Academy of Sciences 17: pp. 650–
655.
———. 1931b. “Proof of the Ergodic Theorem”. Proceedings of the National
Academy of Sciences 17: pp. 656–660.
Blackmore, John. 1999. “Boltzmann and Epistemology”. Synthese 119: pp.
157–189.
Boltzmann, Ludwig. 1872. “Weitere Studien uber das Warmegleichgewicht
unter Gasmolekulen”. Wiener Berichte 66: pp. 275–370.
———. 1877. “Uber die Beziehung zwischen dem zweiten Hauptsatze der
mechanischen Warmetheorie und der Wahrscheinlichkeitsrechnung resp.
118
119
den Satzen uber das Warmegleichgewicht.” Wiener Berichte 76: pp. 373–
435.
———. 1909. Wissenschaftliche Abhandlungen, vol. I, II, and III. Leipzig:
Barth.
Boyling, J. B. 1972. “An Axiomatic Approach to Classical Thermodynam-
ics”. Proceedings of the Royal Society of London. A. Mathematical and
Physical Sciences 329: pp. 35–70.
Bricmont, J. 1995. “Science of Chaos or Chaos in Science?” Annals of the
New York Academy of Sciences 775: pp. 131–175.
———. 2001. “Bayes, Boltzmann and Bohm: Probabilities in Physics”. In
Chance in Physics, edited by J. Bricmont, G. Ghirardi, D. Durr, F. Petruc-
cione, M-C. Galavotti, and N. Zanghı, vol. 574 of Lecture Notes in Physics.
Springer Berlin Heidelberg, pp. 3–21.
Brown, Harvey, Wayne Myrvold, and Jos Uffink. 2009. “Boltzmann’s H-
theorem, its Discontents, and the Birth of Statistical Mechanics”. Studies
in History and Philosophy of Science Part B 40: pp. 174–191.
Brown, Harvey R., and Jos Uffink. 2001. “The Origins of Time-Asymmetry
in Thermodynamics: The Minus First Law”. Studies in History and Phi-
losophy of Science Part B: Studies in History and Philosophy of Modern
Physics 32: pp. 525–538.
Brush, S.G. 1986. The Kind of Motion We Call Heat, vol. 2. North-Holland
Publishing Company.
120
Callender, Craig. 1999. “Reducing Thermodynamics to Statistical Mechan-
ics: The Case of Entropy”. The Journal of Philosophy 96: pp. 348–373.
———. 2001. “Taking Thermodynamics Too Seriously”. Studies in History
and Philosophy of Science Part B: Studies in History and Philosophy of
Modern Physics 32: pp. 539–553.
———. 2010. “The Past Hypothesis Meets Gravity”. In Time, Chance and
Reduction, edited by Gerhard Ernst and Andreas Huttemann. Cambridge
University Press, pp. 34–58.
Caratheodory, C. 1909. “Untersuchungen uber die Grundlagen der Thermo-
dynamik”. Mathematische Annalen 67: pp. 355–386.
Cercignani, C. 1998. Ludwig Boltzmann. The Man who Trusted Atoms. Ox-
ford: Oxford University Press.
Clausius, Rudolf. 1856. “On a Modified Form of the Second Fundamental
Theorem in the Mechanical Theory of Heat”. London, Edinburgh and
Dublin Philosophical Magazine and Journal of Science. Vol. XII: pp.81–
98.
de Grooth, Bart G. 1999. “A Simple Model for Brownian Motion Leading to
the Langevin Equation”. American Journal of Physics 67: pp. 1248–1252.
de Regt, Henk W. 1996. “Philosophy and the Kinetic Theory of Gases”. The
British Journal for the Philosophy of Science 47: pp. 31–62.
Dorfman, Jay Robert. 1999. An Introduction to Chaos in Nonequilibrium
Statistical Mechanics, vol. 14. Cambridge University Press.
121
Earman, John, and Miklos Redei. 1996. “Why Ergodic Theory Does Not
Explain the Success of Equilibrium Statistical Mechanics”. The British
Journal for the Philosophy of Science 47: pp. 63–78.
Ehrenfest, P., and T. Ehrenfest. 2002. The Conceptual Foundations of the
Statistical Approach in Mechanics. Dover Publications.
Ehrenfest-Afanassjewa, T. 1925. “Zur Axiomatisierung des zweiten Haupt-
satzes der Thermodynamik” 33: pp. 933–945.
Einstein, Albert. 1905. “On the Motion—Required by the Molecular Kinetic
Theory of Heat—of Small Particules Suspended in a Stationary Liquid”.
Annalen der Physik 17: pp. 549–560.
Emch, Gerard, and Chuang Liu. 2002. The Logic of Thermo-Statistical
Physics. Springer Berlin Heidelberg.
Fowler, Ralph, and Edward Guggenheim. 1939. Statistical Thermodynamics.
Cambridge: Cambridge University Press.
Friedman, K. S. 1976. “A Partial Vindication of Ergodic Theory”. Philosophy
of Science 43: pp. 151–162.
Frigg, Roman. 2008a. “Chance in Boltzmannian Statistical Mechanics”. Phi-
losophy of Science 75: pp. 670–681.
———. 2008b. “A Field Guide to Recent Work on the Foundations of Statis-
tical Mechanics.” In The Ashgate Companion to Contemporary Philosophy
of Physics, edited by Dean Rickles. Ashgate Publishing.
122
———. 2009. “Typicality and the Approach to Equilibrium in Boltzmannian
Statistical Mechanics”. Philosophy of Science 76: pp. 997–1008.
———. 2011. “Why Typicality Does Not Explain the Approach to Equi-
librium”. In Probabilities, Causes and Propensities in Physics, edited by
M. Suarez, vol. 347. Springer Netherlands, pp. 77–93.
Frigg, Roman, Joseph Berkovitz, and Fred Kronz. 2014. “The Ergodic Hier-
archy”. In The Stanford Encyclopedia of Philosophy, edited by Edward N.
Zalta, summer 2014 edn.
Frigg, Roman, and Charlotte Werndl. 2011. “Explaining Thermodynamic-
Like Behavior in Terms of Epsilon-Ergodicity”. Philosophy of Science 78:
pp. 628–652.
———. 2012a. “Demystifying Typicality”. Philosophy of Science 79: pp.
917–929.
———. 2012b. “A New Approach to the Approach to Equilibrium”. In
Probability in Physics, edited by Y. Ben-Menahem and M. Hemmo. The
Frontiers Collection, Springer Berlin Heidelberg, pp. 99–113.
Giles, Robin. 1964. Mathematical Foundations of Thermodynaics. Oxford:
Pergamon Press.
Goldstein, S. 2001. “Boltzmann’s Approach to Statistical Mechanics”. In
Chance in Physics, edited by J. Bricmont, G. Ghirardi, D. Durr, F. Petruc-
cione, M-C. Galavotti, and N. Zanghı, vol. 574 of Lecture Notes in Physics.
Springer Berlin Heidelberg, pp. 39–54.
123
Goldstein, S., and J. L. Lebowitz. 2004. “On the (Boltzmann) entropy of
non-equilibrium systems”. Physica D: Nonlinear Phenomena 193: pp. 53–
66.
Gottwald, Georg A, and Marcel Oliver. 2009. “Boltzmann’s Dilemma: An
Introduction to Statistical Mechanics via the Kac Ring”. SIAM review 51:
613–635.
Jauch, J. M. 1972. “On a New Foundation of Equilibrium Thermodynamics”.
Foundations of Physics 2: pp. 327–332.
———. 1975. “Analytical Thermodynamics. Part I. Thermostatics—General
theory”. Foundations of Physics 5: pp. 111–132.
Kac, Mark. 1959. Probability and Related Topics in Physical Sciences, vol. 1
of American Mathematical Society. Interscience Publishers.
Kadanoff, Leo. 2000. Statistical Physics: Statics, Dynamics and Renormal-
ization. World Scientific.
Kardar, M. 2007. Statistical Physics of Particles. Cambridge University
Press.
Kelvin. 1851. “On the Dynamical Theory of Heat, with Numerical Results
Deduced from Mr Joule’s Equivalent of a Thermal Unit, and M. Regnault’s
Observations on Steam”. Transactions of the Royal Society of Edinburgh
XX : pp 48–53.
Kestin, J. 1979. A Course in Thermodynamics, vol. Vol. I. New York:
McGraw-Hill.
124
Klein, M. J. 1973. “The Development of Boltzmann’s Statistical Ideas”. In
The Boltzmann Equation: Theory and Applications, edited by D. Cohen
and W. Thirring. Vienna: Springer, pp. 53–106.
Lanford III, O. E. 1975. “Time Evolution of Large Classical Systems”. In
Dynamical Systems, Theory and Applications, edited by J. Moser, vol. 38
of Lecture Notes in Physics. Springer Berlin Heidelberg, pp. 1–111.
———. 1976. “On a Derivation of the Boltzmann Equation”. In Nonequilib-
rium Phenomena I: The Boltzmann Equation, edited by J. L. Lebowitz and
E. W. Montroll, vol. X of Studies In Statistical Mechanics. North-Holland
Publishing Company, pp. 1–17.
———. 1981. “The Hard Sphere Gas in the Boltzmann-Grad Limit”. Physica
A: Statistical Mechanics and its Applications 106: pp. 70–76.
Langevin, Paul. 1908. “Sur la Theorie du Mouvement Brownien”. Comptes
Rendus de l’Acadmie des Sciences 146: pp. 530–533.
Lavis, David A. 2005. “Boltzmann and Gibbs: An Attempted Reconcilia-
tion”. Studies in History and Philosophy of Science Part B: Studies in
History and Philosophy of Modern Physics 36: pp. 245–273.
———. 2008. “Boltzmann, Gibbs, and the Concept of Equilibrium”. Phi-
losophy of Science 75: pp. 682–696.
Lazarovici, Dustin, and Paula Reichert. 2014. “Typicality, Irreversibility and
the Status of Macroscopic Laws”. http://philsci-archive.pitt.edu/10895/.
125
Lebowitz, J. L. 1993a. “Boltzmann’s Entropy and Time’s Arrow”. Physics
Today 46: pp. 32–38.
———. 1993b. “Macroscopic Laws, Microscopic Dynamics, Time’s Arrow
and Boltzmann’s Entropy”. Physica A: Statistical Mechanics and its Ap-
plications 194: pp. 1–27.
———. 1994. “Time’s Arrow and Boltzmann’s Entropy”. In Physical Orgins
of Time Asymmetry, edited by W. H. Zurzek J. J Halliwell, J. Perez-
Mercador. Cambridge University Press.
———. 1999. “Statistical Mechanics: A Selective Review of Two Central
Issues”. Reviews of Modern Physics 71: pp. 346–357.
Lee, John Francis, Francis Weston Sears, and Donald Lawson Turcotte. 1973.
Statistical Mechanics. 2nd edn. Addison Wesley.
Lieb, Elliott H., and Jakob Yngvason. 1999. “The Physics and Mathematics
of the Second Law of Thermodynamics”. Physics Reports 310: pp. 1–96.
Luczak, Joshua, and Wayne C. Myrvold. “Getting Clear About Absolute
Continuity and Translation Continuity”.
Malament, David B., and Sandy L. Zabell. 1980. “Why Gibbs Phase Averages
Work–The Role of Ergodic Theory”. Philosophy of Science 47: pp. 339–
349.
Maxwell, J. C. 1871. Theory of Heat. Longmans, Green, and Company.
Mazenko, Gene. 2006. Nonequilibrium Statistical Mechanics. Wiley.
126
Mendoza, Eric, and Sadi Carnot. 1960. Reflections on the Motive Power of
Fire. Dover Publications.
Myrvold, Wayne C. 2012. “Probabilities in Statistical Mechanics: What are
they?” http://philsci-archive.pitt.edu/9236/.
———. Forthcoming. “Probabilities in Statistical Mechanics”. In Oxford
Handbook of Probability and Philosophy, edited by Christopher Hitchcock
and Alan Hajek. Oxford University Press.
Pathria, Raj, and Paul Beale. 2011. Statistical Mechanics. 3rd edn. Academic
Press.
Pauli, W. 1973. Pauli Lectures on Physics, vol. Volume 3. Thermodynamics
and the Kinetic Theory of Gases. Cambridge, MA: The MIT Press.
Penrose, Roger. 1989. The Emperor’s New Mind. Oxford University Press.
Pitowsky, Itamar. 2012. “Typicality and the Role of the Lebesgue Measure
in Statistical Mechanics”. In Probability in Physics, edited by Yemima
Ben-Menahem and Meir Hemmo. Springer Berlin Heidelberg, pp. 41–58.
Price, Huw. 1996. Time’s Arrow and Archimedes’ Point. Oxford University
Press.
Ramsey, Norman. 1956. “Thermodynamics and Statistical Mechanics at Neg-
ative Absolute Temperatures”. Physical Review 103: pp. 20–28.
Reif, Frederick. 1965. Fundamentals of Statistical and Thermal Physics. New
York: McGraw-Hill.
127
Ridderbos, Katinka. 2002. “The Coarse-Graining Approach to Statistical
Mechanics: How Blissful is Our Ignorance?” Studies in History and Phi-
losophy of Science Part B: Studies in History and Philosophy of Modern
Physics 33: pp. 65–77.
Schroeder, D.V. 2000. An Introduction to Thermal Physics. Addison Wesley.
Schulman, L. S. 1997. Time’s Arrow and Quantum Measurement. Cambridge,
U.K.: Cambridge University Press.
Sklar, L. 1995. Physics and Chance: Philosophical Issues in the Foundations
of Statistical Mechanics. Cambridge University Press.
Smoluchowski, Marian. 1906. “Zur kinetischen Theorie der Brownschen
Molekularbewegung und der Suspensionen”. Annalen der Physik 21: pp.
756–780.
Thompson, Colin J. 1972. Mathematical Statistical Mechanics. The Macmil-
lan Company: New York.
Uffink, Jos. 1996. “Nought But Molecules in Motion”. Studies In History
and Philosophy of Science Part B: Studies In History and Philosophy of
Modern Physics 27: pp. 373–387.
———. 2001. “Bluff Your Way in the Second Law of Thermodynamics”.
Studies in History and Philosophy of Science Part B: Studies in History
and Philosophy of Modern Physics 32: pp. 305–394.
———. 2007. “Compendium of the Foundations of Classical Statistical
128
Physics”. In Philosophy of Phyics, edited by J. Butterfield and J. Ear-
man. Amsterdam: North Holland.
———. 2014. “Boltzmann’s Work in Statistical Physics”. In The Stanford
Encyclopedia of Philosophy, edited by Edward N. Zalta, fall 2014 edn.
Uffink, Jos, and Giovanni Valente. 2015. “Lanford’s Theorem and the Emer-
gence of Irreversibility”. Foundations of Physics 45: pp. 404–438.
Uhlenbeck, G. E., and G. W. Ford. 1963. Lectures in Statistical Mechanics.
Providence: American Mathematical Society.
Valente, Giovanni. 2014. “The Approach Towards Equilibrium in Lanford’s
Theorem”. European Journal for Philosophy of Science : pp. 1–27.
van Lith, J. 2001. “Ergodic Theory, Interpretations of Probability and the
Foundations of Statistical Mechanics”. Studies in History and Philosophy
of Science Part B: Studies in History and Philosophy of Modern Physics
32: pp. 581–594.
Visser, Henk. 1999. “Boltzmann and Wittgenstein or How Pictures Became
Linguistic”. Synthese 119: pp. 135–156.
von Plato, J. 1994. Creating Modern Probability: Its Mathematics, Physics
and Philosophy in Historical Perspective. Cambridge University Press.
Vranas, Peter B. M. 1998. “Epsilon-Ergodicity and the Success of Equilib-
rium Statistical Mechanics”. Philosophy of Science 65: pp. 688–708.
Wallace, David. 2013. “What Statistical Mechanics Actually Does”.
http://philsci-archive.pitt.edu/9846/.
129
Werndl, Charlotte. 2013. “Justifying Typicality Measures of Boltzmannian
Statistical Mechanics and Dynamical Systems”. Studies in History and
Philosophy of Science Part B: Studies in History and Philosophy of Modern
Physics 44: pp. 470–479.
Werndl, Charlotte, and Roman Frigg. 2014. “Rethinking Boltzmannian
Equilibrium”. Forthcoming in: Philosophy of Science. Available at:
http://philsci-archive.pitt.edu/10756/.
———. 2015. “Reconceptualising Equilibrium in Boltzmannian Statistical
Mechanics and Characterising its Existence”. Studies in History and Phi-
losophy of Science Part B: Studies in History and Philosophy of Modern
Physics 49: pp. 19–31.
Zanghı, N. 2005. “I Fondamenti Concettuali dell’Approccio Statistico in
Fisica”. In La Natura Delle Cose. Introduzione ai Fundamenti e alla
Filosofia della Fisica, edited by F. Laudisa V. Allori, M. Dorato and
N. Zanghı. Carocci: Roma.
Zwanzig, Robert. 2001. Nonequilibrium Statistical Mechanics. Oxford Uni-
versity Press.
130
Joshua Luczak
Education
Ph.D., Philosophy 2015The University of Western Ontario
M.A. (First Class Honours), Philosophy 2010Monash University
B.A.\B.Sc. (Hons.), Philosophy, Mathematics 2007Monash University
Publications
2015. “Reflections on Lazare and Sadi Carnot.” Metascience, Vol. 24,Issue 1: 87–89.
2013. “On Chance.” Metascience, Vol. 22, Issue 1: 85–87.
Selected Presentations
“On How to Approach the Approach to Equilibrium.”
15th Congress of Logic, Methodology, and Philosophy of Science. University ofHelsinki, Finland. August 3-8, 2015.
“Toy Models: What They Are, What They’re Not, and What They’reGood For.”
Second Irvine-Pittsburgh-Princeton Conference on the Mathematical and Concep-tual Foundations of Physics. University of California, Irvine, USA. March 20-21,2014.
17th Oxford Philosophy Graduate Conference. University of Oxford, UK. Nov.15-17, 2013.
131
“Toy Models: Interesting? Yes. Approximations? Idealisations? No.”
Models and Simulations 6. University of Notre Dame, USA. May 9-11, 2014.
Congress on Logic and Philosophy of Science. Ghent University, Belgium. Sept.16-18, 2013.
Philosophy Graduate Student Association (PGSA) Colloquium, The Universityof Western Ontario, Canada, March 20, 2013.
“Fatalism, Bivalence, and You”
Canadian Philosophical Association Annual Meeting. Wilfrid Laurier Univer-sity/University of Waterloo, Canada. May 27-30, 2012.
“A Good Defence of Abortion”
Centre on Values and Ethics (COVE) Graduate Student Conference. CarletonUniversity, Canada. March 10-11, 2012.
Teaching
Philosophy Instructor
Georgetown UniversityBeginning Logic Fall 2015Contemporary Moral Issues Fall 2015
Royal Melbourne Institute of TechnologyBusiness Ethics Spring 2010, Fall 2011
Singapore Institute of Management UniversityBusiness Ethics Fall 2011
Philosophy Guest Lecturer
The University of Western OntarioIntroduction to Logic Summer 2013
132
Monash UniversityLife, Death, and Morality Summer 2011Reason and Argument Summer 2010
Philosophy Teaching Assistant1
The University of Western OntarioIntroduction to Philosophy 2013–2014FReasoning and Critical Thinking 2012–2013FAdvanced Introduction to Philosophy 2011–2012F
Monash UniversityHuman Rights Theory Fall 2011Philosophy of Film Summer 2010, Summer 2011Time, Self and Mind Spring 2010Life, Death and Morality Fall 2008, Fall 2009, Fall 2010Philosophical Explorations of World Religions Summer 2010Time, Self and Freedom Spring 2008, Spring 2009God, Freedom and Evil Spring 2009Reason and Argument Fall 2008, Fall 2009Philosophy of Sex Spring 2008
Philosophy Grader
Monash UniversityEthics Summer 2008, Summer 2009Life, Death and Morality Summer 2008Time, Self and Freedom Spring 2007Primary Logic Spring 2007
Mathematics Teaching Assistant
Monash UniversityAnalysis of Change Fall 2006, Fall 2007, Fall 2008Techniques of Modelling Spring 2006, Spring 2008Linear Algebra with Applications Fall 2007Multivariable Calculus Spring 2006
1F = Full year course.
133
Grants, Scholarships, and Awards
Munich Center for Mathematical Philosophy 2016Visiting Fellowship,Ludwig-Maximilians-Universitat
Graduate Thesis Research Award, 2015The University of Western Ontario
Rotman Graduate Research Assistantship, 2013, 2014, 2015The University of Western Ontario
USC Teaching Honour Roll Award of Excellence, 2012-13The University of Western Ontario
British Society for the Philosophy of Science 2013Travel Award,British Society for the Philosophy of Science
Graduate Student Teaching Award Nominee, 2012, 2013The University of Western Ontario
Western Graduate Research Scholarship, 2011–PresentThe University of Western Ontario
Graduate Teaching Assistantship, 2011–PresentThe University of Western Ontario
Faculty of Arts and Humanities 2011Chancellor’s Entrance Scholarship,The University of Western Ontario
Dean’s Commendation of Excellence 2008, 2009, 2010in Teaching,Monash University
Dean’s Sessional Teaching Commendation, 2009Monash University
134
School of Mathematical Science Scholarship, 2006Monash University
Academic Service and Professional Affiliations
PhilPapers Area and Subject Editor, 2014–PresentPhilosophy of Physical Science, Philosophy of ProbabilityThermodynamics and Statistical Mechanics,Models and Idealization, Probabilistic ReasoningPhilpapers.org
Centre for Digital Philosophy 2014–PresentEditorial and Administrative Assistant,The University of Western Ontario
Rotman Institute of Philosophy 2012–PresentStudent Member,The University of Western Ontario
Rotman External Public Outreach 2013–PresentCommittee Member,The University of Western Ontario
Rotman K-12 Outreach 2013–PresentCommittee Member,The University of Western Ontario
Logic, Mathematics, and Physics Graduate 2012, 2013, 2014Philosophy Conference Co-Organiser,The University of Western Ontario
Logic, Mathematics, and Physics Graduate 2013, 2014Philosophy Conference Reviewer,The University of Western Ontario
Philosophy Graduate Student Association Mentor, 2012, 2013The University of Western Ontario